diff --git a/benchmarks/edge_harness/experiments/audit_qwen_tts_real_code_device.py b/benchmarks/edge_harness/experiments/audit_qwen_tts_real_code_device.py new file mode 100644 index 000000000..9e0ee5c64 --- /dev/null +++ b/benchmarks/edge_harness/experiments/audit_qwen_tts_real_code_device.py @@ -0,0 +1,114 @@ +#!/usr/bin/env python3 +"""Bind a generated-code S25 waveform to its pinned TFLite target and CPU references.""" + +from __future__ import annotations + +import argparse +import hashlib +import json +import math +from pathlib import Path + +import numpy as np + + +def sha256(path: Path) -> str: + return hashlib.sha256(path.read_bytes()).hexdigest() + + +def waveform(path: Path, key: str) -> np.ndarray: + with np.load(path, allow_pickle=False) as data: + if data.files != [key]: + raise ValueError(f"{path}: expected only {key}") + value = np.asarray(data[key]) + if value.shape != (1, 48000) or value.dtype != np.float32 or not np.isfinite(value).all(): + raise ValueError(f"{path}: invalid 25-frame waveform {value.shape}/{value.dtype}") + return value + + +def compare(reference: np.ndarray, observed: np.ndarray) -> dict: + a = reference.astype(np.float64).ravel() + b = observed.astype(np.float64).ravel() + error = float(np.linalg.norm(a - b)) + norm = float(np.linalg.norm(a)) + return { + "relative_l2": error / norm, + "snr_db": 20 * math.log10(norm / error) if error else None, + "max_abs": float(np.max(np.abs(a - b))), + "reference_rms": float(np.sqrt(np.mean(a * a))), + "device_rms": float(np.sqrt(np.mean(b * b))), + "device_saturated_fraction_abs_ge_0_999": float(np.mean(np.abs(b) >= 0.999)), + } + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument("--evidence-dir", type=Path, required=True) + parser.add_argument("--pinned-compile", type=Path, required=True) + parser.add_argument("--pinned-export", type=Path, required=True) + args = parser.parse_args() + p = args.evidence_dir + local = json.loads((p / "report.json").read_text()) + dataset = json.loads((p / "dataset_submission.json").read_text()) + submitted = json.loads((p / "inference_submission.json").read_text()) + inference = json.loads((p / "inference_report.json").read_text()) + compiled = json.loads(args.pinned_compile.read_text()) + exported = json.loads(args.pinned_export.read_text()) + if (local["model_revision"] != exported["model_snapshot_revision"] + or local["decoder_weight_sha256"] != exported["model_weight_sha256"] + or local["onnx_sha256"] != exported["onnx_sha256"]): + raise ValueError("generated codes and compiled source are not from the pinned export") + if (local["files"]["fixture"]["sha256"] != dataset["fixture_sha256"] + or dataset["input_shape"] != [1, 512, 97] + or sha256(p / "fixture.npz") != dataset["fixture_sha256"]): + raise ValueError("device input differs from real generated-code fixture") + if (submitted["job_id"] != inference["job_id"] + or inference["status"] != "SUCCESS" + or inference["model_id"] != compiled["target_model_id"] + or inference["input_dataset_id"] != dataset["dataset_id"] + or inference["options"] != "--compute_unit gpu" + or inference["device"] != compiled["device"] + or inference["device"]["name"] != "Samsung Galaxy S25"): + raise ValueError("device result does not bind to pinned S25 GPU target") + for name in ("eager", "ort", "full_reference"): + entry = local["files"][name] + if sha256(p / entry["filename"]) != entry["sha256"]: + raise ValueError(f"local {name} reference changed") + cpu = waveform(p / "ort_output.npz", "wav") + eager = waveform(p / "eager_output.npz", "wav") + full = waveform(p / "full_reference_segment.npz", "wav") + hosted = waveform(p / "device_output.npz", "output_0__0") + comparison = compare(cpu, hosted) + gross_failure = (comparison["relative_l2"] >= 1.0 + or comparison["device_saturated_fraction_abs_ge_0_999"] >= 0.1) + report = { + "scope": "one generated-code 25-frame vocoder window on hosted S25; no device-local talker or complete stream", + "status": "component_gross_numeric_failure" if gross_failure else "component_numeric_measured_no_speech_quality_gate", + "device": inference["device"], + "source_model_id": compiled["source_model_id"], + "target_model_id": compiled["target_model_id"], + "input_dataset_id": dataset["dataset_id"], + "jobs": {"compile": compiled["job_id"], "inference": inference["job_id"]}, + "files": {name: {"sha256": sha256(p / filename), "bytes": (p / filename).stat().st_size} + for name, filename in {"fixture": "fixture.npz", "device": "device_output.npz", + "ort": "ort_output.npz", "eager": "eager_output.npz", + "full": "full_reference_segment.npz"}.items()}, + "ort_cpu_vs_device": comparison, + "eager_cpu_vs_device": compare(eager, hosted), + "full_fp32_decoder_segment_vs_device": compare(full, hosted), + "gross_numeric_failure_screen": { + "failed": gross_failure, + "criteria": "relative L2 >= 1.0 or >=10% saturated samples; not a speech-quality tolerance", + }, + "limits": [ + "One generated-code window is not a speech listening or intelligibility test.", + "Talker, predictor, co-resident handoff, state, cancellation and memory admission did not run on S25.", + "The separate placement profile used Workbench-generated same-shape input, not this fixture.", + ], + } + (p / "audit_report.json").write_text(json.dumps(report, indent=2) + "\n") + print(json.dumps({"status": report["status"], "ort_cpu_vs_device": comparison}, indent=2)) + + +if __name__ == "__main__": + main() diff --git a/benchmarks/edge_harness/experiments/probe_qwen_tts_real_code_window.py b/benchmarks/edge_harness/experiments/probe_qwen_tts_real_code_window.py new file mode 100644 index 000000000..2d1a7c880 --- /dev/null +++ b/benchmarks/edge_harness/experiments/probe_qwen_tts_real_code_window.py @@ -0,0 +1,196 @@ +#!/usr/bin/env python3 +"""Capture one generated CustomVoice code stream and audit a 25-frame export. + +Set PYTHONPATH to include the pinned qwen_tts source checkout when invoking. +The hosted device is not used here; this checks a real-code local window and +its truncation against the full FP32 decoder before a device quality attempt. +""" + +from __future__ import annotations + +import argparse +import hashlib +import inspect +import json +import math +import os +from pathlib import Path + +import numpy as np +import onnxruntime as ort +import torch +from qwen_tts import Qwen3TTSModel +from qwen_tts.core.tokenizer_12hz.configuration_qwen3_tts_tokenizer_v2 import ( + Qwen3TTSTokenizerV2Config, +) +from qwen_tts.core.tokenizer_12hz.modeling_qwen3_tts_tokenizer_v2 import ( + Qwen3TTSTokenizerV2Decoder, +) +from safetensors.torch import load_file + + +def load_decoder(model: Path) -> Qwen3TTSTokenizerV2Decoder: + config = Qwen3TTSTokenizerV2Config.from_pretrained(str(model), subfolder="speech_tokenizer") + decoder = Qwen3TTSTokenizerV2Decoder._from_config(config.decoder_config) + state = load_file(str(model / "speech_tokenizer" / "model.safetensors")) + decoder_state = {key[len("decoder."):]: value for key, value in state.items() + if key.startswith("decoder.")} + missing, unexpected = decoder.load_state_dict(decoder_state, strict=False) + if missing or unexpected: + raise RuntimeError(f"decoder state mismatch: {len(missing)} missing, {len(unexpected)} unexpected") + decoder.eval().float() + if hasattr(decoder, "precompute_snake_caches"): + decoder.precompute_snake_caches() + return decoder + + +def decode_quantized(decoder: Qwen3TTSTokenizerV2Decoder, quantized: torch.Tensor) -> torch.Tensor: + hidden = decoder.pre_conv(quantized).transpose(1, 2) + hidden = decoder.pre_transformer(inputs_embeds=hidden).last_hidden_state + hidden = hidden.permute(0, 2, 1) + for blocks in decoder.upsample: + for block in blocks: + hidden = block(hidden) + for block in decoder.decoder: + hidden = block(hidden) + return hidden.clamp(-1, 1).reshape(1, -1) + + +def sha256(path: Path) -> str: + with path.open("rb") as stream: + return hashlib.file_digest(stream, "sha256").hexdigest() + + +def compare(reference: np.ndarray, observed: np.ndarray) -> dict: + a = reference.astype(np.float64).ravel() + b = observed.astype(np.float64).ravel() + error = float(np.linalg.norm(a - b)) + norm = float(np.linalg.norm(a)) + return { + "relative_l2": error / norm, + "snr_db": 20 * math.log10(norm / error) if error else None, + "max_abs": float(np.max(np.abs(a - b))), + "reference_rms": float(np.sqrt(np.mean(a * a))), + "observed_rms": float(np.sqrt(np.mean(b * b))), + } + + +def main() -> None: + parser = argparse.ArgumentParser(description=__doc__) + parser.add_argument("--model", required=True, type=Path) + parser.add_argument("--onnx", required=True, type=Path) + parser.add_argument("--output-dir", required=True, type=Path) + parser.add_argument("--text", default=( + "The small red square is centered on the white card. I can see its " + "straight edges clearly, and the surrounding background is bright " + "and uncluttered." + )) + parser.add_argument("--max-new-tokens", type=int, default=144) + parser.add_argument("--threads", type=int, default=8) + args = parser.parse_args() + if args.max_new_tokens < 97 or args.threads <= 0: + parser.error("at least 97 max new tokens and positive threads required") + os.environ["HF_HUB_OFFLINE"] = "1" + os.environ["TRANSFORMERS_OFFLINE"] = "1" + torch.set_num_threads(args.threads) + + wrapper = Qwen3TTSModel.from_pretrained( + str(args.model.resolve()), device_map="cpu", dtype=torch.bfloat16, + attn_implementation="sdpa", local_files_only=True, + ) + captured: list[torch.Tensor] = [] + original_decode = wrapper.model.speech_tokenizer.decode + + def capture(items: list[dict], *extra: object, **kwargs: object): + captured.extend(item["audio_codes"].detach().cpu().clone() for item in items) + return original_decode(items, *extra, **kwargs) + + wrapper.model.speech_tokenizer.decode = capture + torch.manual_seed(42) + wavs, sample_rate = wrapper.generate_custom_voice( + text=args.text, language="English", speaker="Ryan", + max_new_tokens=args.max_new_tokens, do_sample=True, + non_streaming_mode=True, + ) + if len(captured) != 1 or len(wavs) != 1 or sample_rate != 24000: + raise RuntimeError("expected one generated code stream and 24 kHz waveform") + codes = captured[0] + if codes.ndim != 2 or codes.shape[0] < 97 or codes.shape[1] != 16: + raise RuntimeError(f"expected >=97 frames of 16-code speech, got {tuple(codes.shape)}") + generated = np.asarray(wavs[0], dtype=np.float32).ravel() + if len(generated) != codes.shape[0] * 1920 or not np.isfinite(generated).all(): + raise RuntimeError("complete request has invalid audio length or samples") + + del wrapper + decoder = load_decoder(args.model) + window_codes = codes[:97].T.unsqueeze(0).contiguous().long() + full_codes = codes.T.unsqueeze(0).contiguous().long() + with torch.inference_mode(): + quantized = decoder.quantizer.decode(window_codes).detach().float().numpy() + eager = decode_quantized(decoder, torch.from_numpy(quantized))[:, -25 * 1920:].detach().numpy() + full = decoder(full_codes).reshape(1, -1).detach().numpy() + if quantized.shape != (1, 512, 97) or eager.shape != (1, 48000): + raise RuntimeError("real-code window violates fixed export shape") + full_segment = full[:, 72 * 1920 : 97 * 1920] + if full_segment.shape != eager.shape: + raise RuntimeError("full decoder output does not align with window") + options = ort.SessionOptions() + options.intra_op_num_threads = args.threads + session = ort.InferenceSession(str(args.onnx), sess_options=options, + providers=["CPUExecutionProvider"]) + observed = np.asarray(session.run(["wav"], {"quantized": quantized})[0]) + if observed.shape != eager.shape or not np.isfinite(observed).all(): + raise RuntimeError("ONNX output violates waveform contract") + + args.output_dir.mkdir(parents=True, exist_ok=True) + files = { + "codes": ("codes.npz", {"codes": codes.numpy()}), + "fixture": ("fixture.npz", {"quantized": quantized}), + "eager": ("eager_output.npz", {"wav": eager}), + "ort": ("ort_output.npz", {"wav": observed}), + "full_reference": ("full_reference_segment.npz", {"wav": full_segment}), + } + for name, (filename, arrays) in files.items(): + np.savez_compressed(args.output_dir / filename, **arrays) + report = { + "scope": "one real generated-code 25-frame Code2Wav window on local CPU; no device execution", + "model_revision": args.model.resolve().name, + "model_weight_sha256": sha256(args.model / "model.safetensors"), + "decoder_weight_sha256": sha256(args.model / "speech_tokenizer" / "model.safetensors"), + "onnx_sha256": sha256(args.onnx), + "torch": torch.__version__, + "onnxruntime": ort.__version__, + "qwen_tts_sources": { + "wrapper": {"path": inspect.getfile(Qwen3TTSModel), + "sha256": sha256(Path(inspect.getfile(Qwen3TTSModel)))}, + "decoder": {"path": inspect.getfile(Qwen3TTSTokenizerV2Decoder), + "sha256": sha256(Path(inspect.getfile(Qwen3TTSTokenizerV2Decoder)))}, + }, + "text": args.text, + "speaker": "Ryan", + "language": "English", + "seed": 42, + "generator_dtype": "bfloat16", + "decoder_dtype": "float32", + "threads": args.threads, + "generated_frames": int(codes.shape[0]), + "generated_audio_samples": len(generated), + "window": {"start_frame": 0, "history_frames": 72, "new_frames": 25}, + "files": {name: {"filename": filename, "sha256": sha256(args.output_dir / filename)} + for name, (filename, _) in files.items()}, + "ort_vs_window_eager": compare(eager, observed), + "window_eager_vs_full_fp32_decoder_segment": compare(full_segment, eager), + "limits": [ + "One generated stream and one window do not establish speech quality.", + "The model generates codes on local CPU; this does not test a mobile talker or handoff.", + "The FP32 full decoder reference is separate from the BF16 generator's waveform.", + ], + } + (args.output_dir / "report.json").write_text(json.dumps(report, indent=2) + "\n") + print(json.dumps({"frames": report["generated_frames"], + "ort_vs_eager": report["ort_vs_window_eager"], + "window_vs_full": report["window_eager_vs_full_fp32_decoder_segment"]}, indent=2)) + + +if __name__ == "__main__": + main() diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/README.md b/benchmarks/edge_harness/results/e2e_expansion_20260924/README.md index d9a670a7a..e17fc8cec 100644 --- a/benchmarks/edge_harness/results/e2e_expansion_20260924/README.md +++ b/benchmarks/edge_harness/results/e2e_expansion_20260924/README.md @@ -34,7 +34,14 @@ from that pinned snapshot matches local eager CPU at 1.99e-6 relative L2 and has compiled to S25 TFLite. Its GPU-requested inference returned an unsaturated waveform at 4.239% relative L2 versus local ONNX CPU, bitwise identical to the historical GPU output on this fixture. Its own placement -profile remains pending; the historical timing is separate evidence. +profile measured 100 component samples at p50/p95 **0.671/1.454 s** for +2 s audio (median RTF **0.335**), with 403 CPU and 335 GPU rows. The +historical profile is a separate run, and neither establishes full streaming. +A [generated-code window](evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/README.md) +from one pinned CustomVoice utterance passed local FP32 window/full-decoder +continuity; the pinned S25 GPU-requested target returned an unsaturated +waveform at **1.153% relative L2 / 38.76 dB SNR** versus local ONNX CPU. +No listening-quality tolerance or complete device-local stream passed. The same pinned [two-frame graph on Galaxy S24](evidence/qwen_tts_vocoder_qualcomm/s24/tflite_short_chunk/README.md) matched local ONNX CPU at 5.75e-6 relative L2 when requested on CPU, while @@ -45,8 +52,10 @@ A FP32-preserving GPU control returned the same saturated waveform bitwise. A [Snapdragon X Elite FP16 QNN short graph](evidence/qwen_tts_vocoder_qualcomm/xelite/qnn_short_chunk/README.md) compiled and returned an unsaturated NPU-requested waveform at 0.930% -relative L2 versus local ONNX CPU on one fixture. Its placement profile -remains pending. These are component-only results. +relative L2 versus local ONNX CPU on one fixture. Its 100-sample profile +attributed all 728 rows to NPU, but p50/p95 was **3.662/3.676 s** for +0.16 s audio (median component RTF **22.89**). These are component-only +results and the short NPU vocoder is too slow for playable streaming. The [Qwen3-TTS vocoder study](evidence/qwen_tts_vocoder_qualcomm/README.md) reuses a retained fixed-shape Code2Wav export and historical Workbench input to test numerical behavior on SA8775P and other Qualcomm settings. The fresh SA8775P GPU-requested inference produces a saturated waveform, while the same binary's CPU-requested output on the exact SA device matches ONNX CPU within 4.13e-6 relative L2. Its 100-sample CPU-requested component profile had p50/p95 5.384/5.534 s with all 738 execution-detail rows on CPU. A FP32-preserving GPU-option control still saturated 84.89% of samples, so that option did not repair the tested path. RB3 GPU-requested execution produces an unsaturated but 15.73%-divergent waveform; its fresh 100-sample mixed CPU/GPU component p50/p95 was 3.099/3.133 s, not a usable TTS timing claim. On the exact RB3 device, the same binary requested on CPU matched original ONNX CPU within 4.13e-6 relative L2 on the retained fixture; its 47-sample component profile measured p50/p95 12.727/13.424 s with all 738 execution-detail rows on CPU. This is a major streaming-latency blocker. A single-fixture-calibrated RB3 INT8 QNN vocoder returned a finite but 66.93%-relative-L2 divergent waveform (3.49 dB SNR) versus ONNX CPU; its 100-sample component profile was p50/p95 1.971/1.992 s with all 728 execution-detail rows on NPU; these timings describe numerically unqualified audio. X Elite ONNX CPU output matches the local ONNX Runtime reference on the retained fixture; its 94-sample component p50/p95 was 6.384/6.644 s with all execution-detail rows on CPU. Corrected X Elite DirectML inference and profile both failed during provider initialization; separate FP16 QNN DLC inference returned an unsaturated waveform at 1.138% relative L2 / 38.88 dB SNR against original ONNX CPU; its placement profile timed out during QNN graph preparation without samples. A separate SA8775P FP16 QNN vocoder compile passed, but same-fixture NPU-requested inference failed after compilation without a waveform or placement samples. Fresh S24 and S25 CPU-requested TFLite vocoder inferences matched ONNX CPU within 4.04e-6 relative L2 on the retained fixture. S25 measured 100 component samples at p50/p95 4.418/4.641 s with all 738 execution-detail rows on CPU; S24 measured 100 at 4.826/4.958 s with all 738 rows on CPU. Separate explicit INT8 and W8A16 QDQ ONNX exports checked on local ORT CPU failed the same-fixture numerical gate at 246.97% and 331.83% waveform relative L2, respectively. S25 CPU talker inference preserved token 80 and its 100-sample component profile measured p50/p95 30.161/31.301 ms with all 1,800 rows on CPU. No complete on-device TTS stream or model-quality qualification is claimed. diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/README.md b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/README.md index da3ef5712..a26112f19 100644 --- a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/README.md +++ b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/README.md @@ -16,14 +16,23 @@ matched its eager decoder at 1.99e-6 relative L2 on the retained synthetic fixture and compiled for S25 TFLite. Its GPU-requested inference returned a finite unsaturated waveform at 4.239% relative L2 versus local ONNX CPU, bitwise identical to the historical target's device output on the same fixture. -Its own placement profile is still running; the historical target's latency -is separate evidence. +Its own 100-sample placement profile measured p50/p95 **0.671/1.454 s** for +2 s audio (median component RTF **0.335**), with 403 CPU and 335 GPU rows. +This is a separate run from the historical target's 0.839/1.157 s profile; +neither is a complete-stream or listening-quality result. +A [generated-code window](s25/real_code_window/README.md) from one pinned +CustomVoice utterance matched local FP32 ONNX/eager and full-decoder segment +within 1.02e-6 and 7.26e-8 relative L2. The same S25 GPU-requested target +returned an unsaturated waveform at **1.153% relative L2 / 38.76 dB SNR** +versus ONNX CPU on that real-code window. This has no listening-quality or +full-pipeline gate. The same pinned two-frame source [compiled for X Elite QNN](xelite/qnn_short_chunk/README.md) and its NPU-requested inference returned an unsaturated waveform at 0.930% -relative L2 / 40.63 dB SNR versus local ONNX CPU. Its placement profile is -still running; this is component numerical evidence, not verified NPU node -placement or complete TTS. +relative L2 / 40.63 dB SNR versus local ONNX CPU. Its 100-sample placement +profile attributed all 728 rows to NPU, but p50/p95 was **3.662/3.676 s** +for only 0.16 s audio (median component RTF **22.89**). This is a measured +playable-stream blocker for that artifact, not complete TTS. On [Galaxy S24](s24/tflite_short_chunk/README.md), the same pinned two-frame TFLite source matched local ONNX CPU at 5.75e-6 relative L2 when requested on CPU, but its GPU-requested output saturated every sample and differed by 24.218 relative L2. The 100-sample CPU component p50/p95 was 3.433/3.873 s for only 0.16 s audio (median RTF 21.46), with all 738 rows on CPU. The mixed CPU/GPU component measured 0.601/0.895 s with 403 CPU and 335 GPU rows, but that timing describes numerically failed audio. An [FP32-preserving delegate control](s24/tflite_short_chunk_gpu_fp32/audit_report.json) returned the same saturated waveform bitwise, so that option did not fix this artifact. Neither route is a complete S24 TTS stream. diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/full_pinned_tflite/README.md b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/full_pinned_tflite/README.md index bac3e8bac..522a4d234 100644 --- a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/full_pinned_tflite/README.md +++ b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/full_pinned_tflite/README.md @@ -20,6 +20,18 @@ The new device waveform is [bitwise identical](historical_device_parity_report.j to the [historical 25-frame target's](../full_tflite_gpu_profile/README.md) GPU output on this retained fixture. This supports numerical continuity on one input, not identical historical export provenance or listening quality. -A separate [placement profile](profile_submission.json) is in progress; GPU -node placement and latency for this revision-pinned target remain unverified. -Complete-stream handoff and sustained behavior are also unverified. +A separate [placement profile](profile_report.json) returned 100 component +samples at nearest-rank p50/p95 **0.671/1.454 s** for 2 s audio, with 403 CPU +and 335 GPU execution-detail rows and 380,100,608 bytes reported peak memory. +The median component RTF is **0.335**. Workbench generated profile input of +the same shape, while the numerical inference used the retained fixture; the +[audit](audit_report.json) binds both jobs to the exact target. Complete-stream +handoff, listening quality and sustained behavior remain unverified. + +A separate [real generated-code window](../real_code_window/README.md) passed +local ONNX/eager parity at 1.02e-6 relative L2 and matched the corresponding +full FP32 decode segment at 7.26e-8 relative L2. The same fixture has a +successful inference on this pinned S25 target: the unsaturated device +waveform differed from local ONNX CPU by 1.153% relative L2 / 38.76 dB SNR. +This one real-code window is encouraging component evidence, not a speech +quality tolerance or complete mobile stream. diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/full_pinned_tflite/audit_report.json b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/full_pinned_tflite/audit_report.json index 0c203d2ac..c5234153a 100644 --- a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/full_pinned_tflite/audit_report.json +++ b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/full_pinned_tflite/audit_report.json @@ -1,6 +1,6 @@ { "scope": "one 25-frame Qwen3-TTS vocoder component on hosted Samsung Galaxy S25; no complete stream", - "status": "component_inference_numeric_measured", + "status": "component_inference_numeric_and_placement_measured", "device": { "name": "Samsung Galaxy S25", "os": "15", @@ -25,7 +25,8 @@ "input_dataset_id": "d2nmpxd57", "jobs": { "compile": "jgddd1zzg", - "inference": "jglyymx25" + "inference": "jglyymx25", + "profile": "j568847ng" }, "route": { "compile_options": "--target_runtime tflite", @@ -72,7 +73,18 @@ "limits": [ "One fixed-window waveform has no listening-quality tolerance.", "The 72-frame input history, talker, predictor and stage handoff did not run on the device.", - "No persistent state, admission, cancellation or complete TTS stream was tested.", - "Requested accelerator execution is not verified node placement without a profile." - ] + "No persistent state, admission, cancellation or complete TTS stream was tested." + ], + "profile": { + "sample_count": 100, + "min_us": 560210, + "nearest_rank_p50_us": 670776, + "nearest_rank_p95_us": 1453828, + "reported_peak_memory_bytes": 380100608, + "compute_unit_row_counts": { + "CPU": 403, + "GPU": 335 + }, + "profile_input": "Workbench generated input with audited tensor shape; the waveform inference used the retained fixture" + } } diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/full_pinned_tflite/profile_report.json b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/full_pinned_tflite/profile_report.json new file mode 100644 index 000000000..488563f80 --- /dev/null +++ b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/full_pinned_tflite/profile_report.json @@ -0,0 +1,5362 @@ +{ + "timestamp_utc": "2026-09-24T06:46:29.252274+00:00", + "qai_hub_version": "0.55.0", + "facility": "Qualcomm AI Hub Workbench", + "job_id": "j568847ng", + "url": "https://workbench.aihub.qualcomm.com/jobs/j568847ng/", + "status": "SUCCESS", + "status_message": "", + "options": "--compute_unit gpu", + "hub_version": "aihub-2026.09.11.0", + "device": { + "name": "Samsung Galaxy S25", + "os": "15", + "attributes": [ + "os:android", + "framework:tflite", + "framework:onnx", + "abi:aarch64-android", + "vendor:samsung", + "format:phone", + "chipset:qualcomm-snapdragon-8-elite-for-galaxy", + "chipset:sm8750-ac", + "hexagon:v79", + "soc-model:69", + "htp-supports-fp16:true", + "htp-supports-weight-sharing:true", + "framework:qnn" + ] + }, + "model_id": "mn4o5j17q", + "shapes": { + "quantized": [ + [ + 1, + 512, + 97 + ], + "float32" + ] + }, + "profile": { + "execution_summary": { + "estimated_inference_time": 560210, + "estimated_inference_peak_memory": 380100608, + "first_load_time": 63813836, + "first_load_peak_memory": 824541184, + "warm_load_time": 342237, + "warm_load_peak_memory": 650473472, + "compile_time": 0, + "compile_peak_memory": 0, + "compile_memory_increase_range": [ + 0, + 0 + ], + "compile_memory_peak_range": [ + 0, + 0 + ], + "first_load_memory_increase_range": [ + 276747904, + 285359344 + ], + "first_load_memory_peak_range": [ + 725114880, + 733726320 + ], + "warm_load_memory_increase_range": [ + 252154432, + 274439296 + ], + "warm_load_memory_peak_range": [ + 555655168, + 577940032 + ], + "inference_memory_increase_range": [ + 0, + 1039920 + ], + "inference_memory_peak_range": [ + 1261568, + 33479936 + ], + "all_compile_times": [], + "all_first_load_times": [ + 63813836 + ], + "all_warm_load_times": [ + 342237 + ], + "all_inference_times": [ + 560210, + 581072, + 574676, + 599914, + 603034, + 619793, + 798962, + 832110, + 857496, + 831400, + 829756, + 794990, + 858877, + 831684, + 821768, + 850296, + 822182, + 837954, + 811579, + 836368, + 865660, + 851235, + 834303, + 846860, + 932888, + 1494083, + 1469218, + 1490056, + 1453828, + 1479436, + 1472690, + 1332117, + 1158977, + 1066258, + 988709, + 985714, + 1024498, + 1144765, + 1167878, + 1186269, + 1030869, + 892563, + 856154, + 680146, + 645514, + 636818, + 643249, + 641513, + 642717, + 631295, + 633373, + 619866, + 641588, + 642864, + 646676, + 635843, + 640958, + 631941, + 667637, + 645856, + 643275, + 639485, + 647008, + 644643, + 638704, + 647272, + 640466, + 639873, + 647446, + 653926, + 643621, + 653569, + 643166, + 643910, + 653790, + 636806, + 670776, + 642877, + 644992, + 648707, + 669081, + 645564, + 641579, + 647462, + 647751, + 645482, + 663250, + 663109, + 674582, + 705578, + 722180, + 693089, + 703415, + 715765, + 700714, + 700429, + 699122, + 706386, + 695987, + 702627 + ] + }, + "execution_detail": [ + { + "name": "model/tf.compat.v1.transpose/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 41, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.pad/Pad", + "type": "PAD", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "CPU", + "execution_time": 2032, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_1/Add", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 628, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean/Mean", + "type": "MEAN", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_2/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sqrt/Sqrt", + "type": "SQRT", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reciprocal/Reciprocal", + "type": "DIV", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_1/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 1645, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_1/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 10, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_1/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 9, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_2/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_3/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_4/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.negative/Neg", + "type": "NEG", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.concat/node_cat_3", + "type": "CONCATENATION", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_3/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_3/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_1/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_1/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_2/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_4/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_5/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_6/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.negative_1/Neg", + "type": "NEG", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.concat_1/node_cat_4", + "type": "CONCATENATION", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_5/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_4/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 115, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_4/Add", + "type": "BATCH_MATMUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_6/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 8, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_5/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 49, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.softmax/Softmax", + "type": "SOFTMAX", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_2/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_2/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 17, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_3/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 116, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_3/MatMul", + "type": "BATCH_MATMUL", + "compute_unit": "CPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_5/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_3/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 509, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_7/Mul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 3, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_6/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 1, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_8/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_1/Mean", + "type": "MEAN", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_7/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sqrt_1/Sqrt", + "type": "SQRT", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reciprocal_1/Reciprocal", + "type": "DIV", + "compute_unit": "CPU", + "execution_time": 519, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_9/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 30, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_5/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sigmoid/Sigmoid", + "type": "LOGISTIC", + "compute_unit": "CPU", + "execution_time": 497, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_11/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_6/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 514, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_12/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 3, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_13/Mul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 1, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_8/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_14/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_2/Mean", + "type": "MEAN", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_9/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sqrt_2/Sqrt", + "type": "SQRT", + "compute_unit": "CPU", + "execution_time": 1469, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reciprocal_2/Reciprocal", + "type": "DIV", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_15/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_8/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 8, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_7/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 8, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_4/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_6/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_16/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_10/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_11/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.negative_2/Neg", + "type": "NEG", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.concat_2/node_cat_5", + "type": "CONCATENATION", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_17/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_10/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_8/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_5/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_7/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_18/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 1, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_12/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_13/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.negative_3/Neg", + "type": "NEG", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.concat_3/node_cat_6", + "type": "CONCATENATION", + "compute_unit": "CPU", + "execution_time": 108, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_19/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_11/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 8, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_11/Add", + "type": "BATCH_MATMUL", + "compute_unit": "CPU", + "execution_time": 48, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_20/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_12/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.softmax_1/Softmax", + "type": "SOFTMAX", + "compute_unit": "CPU", + "execution_time": 16, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_9/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 94, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_6/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_8/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_10/MatMul", + "type": "BATCH_MATMUL", + "compute_unit": "CPU", + "execution_time": 389, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_10/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 3, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_7/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 1, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_21/Mul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_13/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_22/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_3/Mean", + "type": "MEAN", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_14/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 455, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sqrt_3/Sqrt", + "type": "SQRT", + "compute_unit": "CPU", + "execution_time": 29, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reciprocal_3/Reciprocal", + "type": "DIV", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_23/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 458, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_12/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sigmoid_1/Sigmoid", + "type": "LOGISTIC", + "compute_unit": "CPU", + "execution_time": 498, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_25/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 3, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_13/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 1, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_26/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_27/Mul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_15/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_28/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_4/Mean", + "type": "MEAN", + "compute_unit": "CPU", + "execution_time": 1060, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_16/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sqrt_4/Sqrt", + "type": "SQRT", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reciprocal_4/Reciprocal", + "type": "DIV", + "compute_unit": "CPU", + "execution_time": 7, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_29/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 7, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_15/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_14/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_8/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_11/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_30/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_17/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_18/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.negative_4/Neg", + "type": "NEG", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.concat_4/node_cat_7", + "type": "CONCATENATION", + "compute_unit": "CPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_31/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_17/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_15/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_9/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_12/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_32/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_19/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_20/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 85, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.negative_5/Neg", + "type": "NEG", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.concat_5/node_cat_8", + "type": "CONCATENATION", + "compute_unit": "CPU", + "execution_time": 8, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_33/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 48, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_18/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_18/Add", + "type": "BATCH_MATMUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_34/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 14, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_19/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 74, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.softmax_2/Softmax", + "type": "SOFTMAX", + "compute_unit": "CPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_16/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_10/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 353, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_13/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 3, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_17/MatMul", + "type": "BATCH_MATMUL", + "compute_unit": "CPU", + "execution_time": 1, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_15/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_11/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_35/Mul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_20/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_36/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 365, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_5/Mean", + "type": "MEAN", + "compute_unit": "CPU", + "execution_time": 21, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_21/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sqrt_5/Sqrt", + "type": "SQRT", + "compute_unit": "CPU", + "execution_time": 359, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reciprocal_5/Reciprocal", + "type": "DIV", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_37/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 349, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_19/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 3, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sigmoid_2/Sigmoid", + "type": "LOGISTIC", + "compute_unit": "CPU", + "execution_time": 1, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_39/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_20/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_40/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_41/Mul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_22/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 1005, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_42/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_6/Mean", + "type": "MEAN", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_23/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 7, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sqrt_6/Sqrt", + "type": "SQRT", + "compute_unit": "CPU", + "execution_time": 7, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reciprocal_6/Reciprocal", + "type": "DIV", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_43/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_22/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_21/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_12/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_16/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_44/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_24/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_25/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.negative_6/Neg", + "type": "NEG", + "compute_unit": "CPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.concat_6/node_cat_9", + "type": "CONCATENATION", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_45/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_24/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_22/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_13/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_17/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_46/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 86, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_26/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_27/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 7, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.negative_7/Neg", + "type": "NEG", + "compute_unit": "CPU", + "execution_time": 48, + "execution_cycles": 0 + }, + { + "name": "model/tf.concat_7/node_cat_10", + "type": "CONCATENATION", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_47/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_25/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 14, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_25/Add", + "type": "BATCH_MATMUL", + "compute_unit": "CPU", + "execution_time": 71, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_48/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_26/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.softmax_3/Softmax", + "type": "SOFTMAX", + "compute_unit": "CPU", + "execution_time": 322, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_23/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 3, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_14/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 1, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_18/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_24/MatMul", + "type": "BATCH_MATMUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_20/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_15/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_49/Mul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 329, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_27/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 20, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_50/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_7/Mean", + "type": "MEAN", + "compute_unit": "CPU", + "execution_time": 335, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_28/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sqrt_7/Sqrt", + "type": "SQRT", + "compute_unit": "CPU", + "execution_time": 343, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reciprocal_7/Reciprocal", + "type": "DIV", + "compute_unit": "CPU", + "execution_time": 3, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_51/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 1, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_26/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sigmoid_3/Sigmoid", + "type": "LOGISTIC", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_53/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_27/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_54/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 966, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_55/Mul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_29/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_56/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 7, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_8/Mean", + "type": "MEAN", + "compute_unit": "CPU", + "execution_time": 7, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_30/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sqrt_8/Sqrt", + "type": "SQRT", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reciprocal_8/Reciprocal", + "type": "DIV", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_57/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_29/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_28/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_16/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_21/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_58/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_31/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_32/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.negative_8/Neg", + "type": "NEG", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.concat_8/node_cat_11", + "type": "CONCATENATION", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_59/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_31/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_29/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_17/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 80, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_22/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_60/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 8, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_33/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 45, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_34/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.negative_9/Neg", + "type": "NEG", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.concat_9/node_cat_12", + "type": "CONCATENATION", + "compute_unit": "CPU", + "execution_time": 13, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_61/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 68, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_32/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_32/Add", + "type": "BATCH_MATMUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_62/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 334, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_33/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 3, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.softmax_4/Softmax", + "type": "SOFTMAX", + "compute_unit": "CPU", + "execution_time": 1, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_30/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_18/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_23/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_31/MatMul", + "type": "BATCH_MATMUL", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_25/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 330, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_19/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 20, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_63/Mul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_34/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 331, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_64/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_9/Mean", + "type": "MEAN", + "compute_unit": "CPU", + "execution_time": 324, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_35/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 3, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sqrt_9/Sqrt", + "type": "SQRT", + "compute_unit": "CPU", + "execution_time": 1, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reciprocal_9/Reciprocal", + "type": "DIV", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_65/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_33/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sigmoid_4/Sigmoid", + "type": "LOGISTIC", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_67/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 934, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_34/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_68/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_69/Mul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 7, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_36/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 7, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_70/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_10/Mean", + "type": "MEAN", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_37/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sqrt_10/Sqrt", + "type": "SQRT", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reciprocal_10/Reciprocal", + "type": "DIV", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_71/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_36/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_35/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_20/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_26/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_72/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_38/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_39/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.negative_10/Neg", + "type": "NEG", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.concat_10/node_cat_13", + "type": "CONCATENATION", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_73/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_38/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 79, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_36/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_21/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 7, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_27/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 45, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_74/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_40/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_41/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 13, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.negative_11/Neg", + "type": "NEG", + "compute_unit": "CPU", + "execution_time": 68, + "execution_cycles": 0 + }, + { + "name": "model/tf.concat_11/node_cat_14", + "type": "CONCATENATION", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_75/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_39/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 321, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_39/Add", + "type": "BATCH_MATMUL", + "compute_unit": "CPU", + "execution_time": 3, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_76/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 1, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_40/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.softmax_5/Softmax", + "type": "SOFTMAX", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_37/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_22/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_28/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 329, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_38/MatMul", + "type": "BATCH_MATMUL", + "compute_unit": "CPU", + "execution_time": 19, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_30/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_23/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 329, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_77/Mul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_41/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 322, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_78/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_11/Mean", + "type": "MEAN", + "compute_unit": "CPU", + "execution_time": 1, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_42/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sqrt_11/Sqrt", + "type": "SQRT", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reciprocal_11/Reciprocal", + "type": "DIV", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_79/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_40/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 934, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sigmoid_5/Sigmoid", + "type": "LOGISTIC", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_81/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_41/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 7, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_82/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 7, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_83/Mul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_43/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_84/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_12/Mean", + "type": "MEAN", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_44/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sqrt_12/Sqrt", + "type": "SQRT", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reciprocal_12/Reciprocal", + "type": "DIV", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_85/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_43/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_42/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_24/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_31/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_86/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_45/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_46/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.negative_12/Neg", + "type": "NEG", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.concat_12/node_cat_15", + "type": "CONCATENATION", + "compute_unit": "CPU", + "execution_time": 79, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_87/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_45/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 8, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_43/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 45, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_25/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_32/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_88/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 12, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_47/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 68, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_48/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.negative_13/Neg", + "type": "NEG", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.concat_13/node_cat_16", + "type": "CONCATENATION", + "compute_unit": "CPU", + "execution_time": 322, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_89/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 3, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_46/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 1, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_46/Add", + "type": "BATCH_MATMUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_90/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_47/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.softmax_6/Softmax", + "type": "SOFTMAX", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_44/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 332, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_26/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 20, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_33/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_45/MatMul", + "type": "BATCH_MATMUL", + "compute_unit": "CPU", + "execution_time": 328, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_35/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_27/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 321, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_91/Mul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_48/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 1, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_92/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_13/Mean", + "type": "MEAN", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_49/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sqrt_13/Sqrt", + "type": "SQRT", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reciprocal_13/Reciprocal", + "type": "DIV", + "compute_unit": "CPU", + "execution_time": 931, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_93/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_47/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sigmoid_6/Sigmoid", + "type": "LOGISTIC", + "compute_unit": "CPU", + "execution_time": 7, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_95/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 7, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_48/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_96/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_97/Mul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_50/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_98/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_14/Mean", + "type": "MEAN", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_51/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sqrt_14/Sqrt", + "type": "SQRT", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reciprocal_14/Reciprocal", + "type": "DIV", + "compute_unit": "CPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_99/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_50/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_49/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_28/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_36/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_100/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_52/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_53/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 79, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.negative_14/Neg", + "type": "NEG", + "compute_unit": "CPU", + "execution_time": 5, + "execution_cycles": 0 + }, + { + "name": "model/tf.concat_14/node_cat_17", + "type": "CONCATENATION", + "compute_unit": "CPU", + "execution_time": 12, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_101/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_52/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_50/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 10, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_29/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_37/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_102/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_54/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_55/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.negative_15/Neg", + "type": "NEG", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.concat_15/node_cat_18", + "type": "CONCATENATION", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_103/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_53/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_53/Add", + "type": "BATCH_MATMUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_104/Mul", + "type": "MUL", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_54/Add", + "type": "ADD", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.softmax_7/Softmax", + "type": "SOFTMAX", + "compute_unit": "GPU", + "execution_time": 30, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_51/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_30/Reshape", + "type": "RESHAPE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_38/transpose", + "type": "TRANSPOSE", + "compute_unit": "CPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_52/MatMul", + "type": "BATCH_MATMUL", + "compute_unit": "GPU", + "execution_time": 30, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_40/transpose", + "type": "TRANSPOSE", + "compute_unit": "GPU", + "execution_time": 3, + "execution_cycles": 0 + }, + { + "name": "model/tf.reshape_31/Reshape", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_105/Mul", + "type": "FULLY_CONNECTED", + "compute_unit": "GPU", + "execution_time": 166, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_55/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_106/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_15/Mean", + "type": "MEAN", + "compute_unit": "GPU", + "execution_time": 28, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_56/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sqrt_15/Sqrt", + "type": "SQRT", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reciprocal_15/Reciprocal", + "type": "DIV", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_107/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 3, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_54/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "GPU", + "execution_time": 128, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sigmoid_7/Sigmoid", + "type": "LOGISTIC", + "compute_unit": "GPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_109/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.linalg.matmul_55/MatMul", + "type": "FULLY_CONNECTED", + "compute_unit": "GPU", + "execution_time": 126, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_110/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_111/Mul", + "type": "FULLY_CONNECTED", + "compute_unit": "GPU", + "execution_time": 164, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_57/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_112/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_16/Mean", + "type": "MEAN", + "compute_unit": "GPU", + "execution_time": 29, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_58/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sqrt_16/Sqrt", + "type": "SQRT", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reciprocal_16/Reciprocal", + "type": "DIV", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_113/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 2, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_59/Add", + "type": "FULLY_CONNECTED", + "compute_unit": "GPU", + "execution_time": 126, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.conv1d_transpose/conv1d_transpose/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.conv1d_transpose/conv1d_transpose", + "type": "TRANSPOSE_CONV", + "compute_unit": "GPU", + "execution_time": 449, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.conv1d_transpose/conv1d_transpose/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.pad_1/Pad", + "type": "PAD", + "compute_unit": "GPU", + "execution_time": 7, + "execution_cycles": 0 + }, + { + "name": "model/node_conv1d_1/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/node_conv1d_1/depthwise", + "type": "DEPTHWISE_CONV_2D", + "compute_unit": "GPU", + "execution_time": 16, + "execution_cycles": 0 + }, + { + "name": "model/node_conv1d_1/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_17/Mean", + "type": "MEAN", + "compute_unit": "GPU", + "execution_time": 8, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.squared_difference/SquaredDifference", + "type": "SQUARED_DIFFERENCE", + "compute_unit": "GPU", + "execution_time": 4, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_18/Mean", + "type": "MEAN", + "compute_unit": "GPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.__operators__.add/AddV2", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.rsqrt/Rsqrt", + "type": "RSQRT", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_114/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 3, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_115/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 13, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_116/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 11, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.subtract/Sub", + "type": "SUB", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.__operators__.add_1/AddV2", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_62/Add", + "type": "FULLY_CONNECTED", + "compute_unit": "GPU", + "execution_time": 1866, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.gelu/Gelu/Erfc", + "type": "GELU", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_63/Add", + "type": "FULLY_CONNECTED", + "compute_unit": "GPU", + "execution_time": 2102, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_64/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.conv1d_transpose_1/conv1d_transpose/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.conv1d_transpose_1/conv1d_transpose", + "type": "TRANSPOSE_CONV", + "compute_unit": "GPU", + "execution_time": 935, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.conv1d_transpose_1/conv1d_transpose/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.pad_2/Pad", + "type": "PAD", + "compute_unit": "GPU", + "execution_time": 14, + "execution_cycles": 0 + }, + { + "name": "model/node_conv1d_2/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/node_conv1d_2/depthwise", + "type": "DEPTHWISE_CONV_2D", + "compute_unit": "GPU", + "execution_time": 23, + "execution_cycles": 0 + }, + { + "name": "model/node_conv1d_2/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_19/Mean", + "type": "MEAN", + "compute_unit": "GPU", + "execution_time": 13, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.squared_difference_1/SquaredDifference", + "type": "SQUARED_DIFFERENCE", + "compute_unit": "GPU", + "execution_time": 14, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.reduce_mean_20/Mean", + "type": "MEAN", + "compute_unit": "GPU", + "execution_time": 13, + "execution_cycles": 0 + }, + { + "name": "model/tf.__operators__.add_2/AddV2", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.rsqrt_1/Rsqrt", + "type": "RSQRT", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_117/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 6, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_118/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 20, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_119/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 23, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.subtract_1/Sub", + "type": "SUB", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.__operators__.add_3/AddV2", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_67/Add", + "type": "FULLY_CONNECTED", + "compute_unit": "GPU", + "execution_time": 3181, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.gelu_1/Gelu/Erfc", + "type": "GELU", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_68/Add", + "type": "FULLY_CONNECTED", + "compute_unit": "GPU", + "execution_time": 3631, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_69/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.pad_3/Pad", + "type": "PAD", + "compute_unit": "GPU", + "execution_time": 15, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_1/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_1/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 9351, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_1/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_120/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 22, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_121/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_122/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_71/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.conv1d_transpose_2/conv1d_transpose/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.conv1d_transpose_2/conv1d_transpose", + "type": "TRANSPOSE_CONV", + "compute_unit": "GPU", + "execution_time": 34374, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.conv1d_transpose_2/conv1d_transpose/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_56/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "GPU", + "execution_time": 131, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_123/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 133, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_1/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_124/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_125/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_73/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.pad_4/Pad", + "type": "PAD", + "compute_unit": "GPU", + "execution_time": 133, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_2/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_2/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 24504, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_2/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_126/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 128, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_2/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_127/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_128/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_75/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_3/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_3/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 3525, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_3/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_77/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_129/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 130, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_3/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_130/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_131/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_78/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.pad_5/Pad", + "type": "PAD", + "compute_unit": "GPU", + "execution_time": 141, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_4/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_4/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 24511, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_4/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_132/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 130, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_4/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_133/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_134/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_80/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_5/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_5/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 3526, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_5/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_82/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_135/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 129, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_5/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_136/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_137/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_83/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.pad_6/Pad", + "type": "PAD", + "compute_unit": "GPU", + "execution_time": 147, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_6/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_6/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 24524, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_6/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_138/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 129, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_6/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_139/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_140/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_85/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_7/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_7/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 3525, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_7/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_87/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_141/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 129, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_7/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_142/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_143/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_88/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.conv1d_transpose_3/conv1d_transpose/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.conv1d_transpose_3/conv1d_transpose", + "type": "TRANSPOSE_CONV", + "compute_unit": "GPU", + "execution_time": 49712, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.conv1d_transpose_3/conv1d_transpose/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_57/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "GPU", + "execution_time": 337, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_144/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 345, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_8/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_145/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_146/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_90/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.pad_7/Pad", + "type": "PAD", + "compute_unit": "GPU", + "execution_time": 364, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_8/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_8/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 30454, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_8/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_147/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 337, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_9/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_148/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_149/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_92/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_9/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_9/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 4388, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_9/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_94/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_150/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 342, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_10/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_151/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_152/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_95/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.pad_8/Pad", + "type": "PAD", + "compute_unit": "GPU", + "execution_time": 358, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_10/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_10/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 30949, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_10/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_153/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 336, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_11/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_154/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_155/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_97/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_11/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_11/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 4386, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_11/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_99/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_156/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 334, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_12/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_157/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_158/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_100/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.pad_9/Pad", + "type": "PAD", + "compute_unit": "GPU", + "execution_time": 361, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_12/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_12/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 30514, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_12/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_159/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 337, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_13/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_160/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_161/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_102/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_13/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_13/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 4386, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_13/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_104/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_162/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 342, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_14/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_163/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_164/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_105/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.conv1d_transpose_4/conv1d_transpose/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.conv1d_transpose_4/conv1d_transpose", + "type": "TRANSPOSE_CONV", + "compute_unit": "GPU", + "execution_time": 26049, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.conv1d_transpose_4/conv1d_transpose/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_58/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "GPU", + "execution_time": 685, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_165/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 684, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_15/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_166/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_167/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_107/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.pad_10/Pad", + "type": "PAD", + "compute_unit": "GPU", + "execution_time": 707, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_14/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_14/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 34403, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_14/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_168/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 687, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_16/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_169/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_170/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_109/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_15/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_15/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 4788, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_15/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_111/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_171/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 691, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_17/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_172/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_173/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_112/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.pad_11/Pad", + "type": "PAD", + "compute_unit": "GPU", + "execution_time": 697, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_16/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_16/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 34422, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_16/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_174/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 685, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_18/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_175/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_176/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_114/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_17/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_17/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 4792, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_17/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_116/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_177/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 686, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_19/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_178/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_179/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_117/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.pad_12/Pad", + "type": "PAD", + "compute_unit": "GPU", + "execution_time": 695, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_18/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_18/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 34404, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_18/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_180/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 687, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_20/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_181/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_182/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_119/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_19/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_19/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 4788, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_19/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_121/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_183/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 691, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_21/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_184/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_185/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_122/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.conv1d_transpose_5/conv1d_transpose/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.conv1d_transpose_5/conv1d_transpose", + "type": "TRANSPOSE_CONV", + "compute_unit": "GPU", + "execution_time": 14693, + "execution_cycles": 0 + }, + { + "name": "model/tf.nn.conv1d_transpose_5/conv1d_transpose/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_59/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "GPU", + "execution_time": 1049, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_186/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 1048, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_22/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_187/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_188/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_124/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.pad_13/Pad", + "type": "PAD", + "compute_unit": "GPU", + "execution_time": 1057, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_20/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_20/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 25828, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_20/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_189/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 1037, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_23/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_190/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_191/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_126/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_21/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_21/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 3664, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_21/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_128/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_192/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 1042, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_24/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_193/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_194/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_129/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.pad_14/Pad", + "type": "PAD", + "compute_unit": "GPU", + "execution_time": 1036, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_22/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_22/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 25824, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_22/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_195/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 1036, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_25/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_196/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_197/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_131/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_23/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_23/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 3664, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_23/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_133/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_198/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 1041, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_26/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_199/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_200/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_134/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.pad_15/Pad", + "type": "PAD", + "compute_unit": "GPU", + "execution_time": 1047, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_24/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_24/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 25826, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_24/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_201/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 1037, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_27/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_202/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_203/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_136/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_25/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_25/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 3640, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_25/DynamicConv1D/Squeeze", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_138/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_204/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 1041, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.sin_28/Sin", + "type": "SIN", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_205/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.multiply_206/Mul", + "type": "MUL", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.math.add_139/Add", + "type": "ADD", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.pad_16/Pad", + "type": "PAD", + "compute_unit": "GPU", + "execution_time": 1056, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_26/DynamicConv1D/ExpandDims", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "model/dynamic_conv1d_26/DynamicConv1D", + "type": "CONV_2D", + "compute_unit": "GPU", + "execution_time": 602, + "execution_cycles": 0 + }, + { + "name": "model/tf.compat.v1.transpose_41/transpose", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 29, + "execution_cycles": 0 + }, + { + "name": "model/tf.strided_slice_60/StridedSlice", + "type": "STRIDED_SLICE", + "compute_unit": "GPU", + "execution_time": 7, + "execution_cycles": 0 + }, + { + "name": "output_0", + "type": "RESHAPE", + "compute_unit": "GPU", + "execution_time": 20, + "execution_cycles": 0 + } + ] + } +} diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/README.md b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/README.md new file mode 100644 index 000000000..ec0a7a8da --- /dev/null +++ b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/README.md @@ -0,0 +1,27 @@ +# S25 pinned vocoder on generated speech codes + +The [reproducer](../../../../../../experiments/probe_qwen_tts_real_code_window.py) +ran the pinned Qwen3-TTS 0.6B CustomVoice checkpoint at revision +`85e237c12c027371202489a0ec509ded67b5e4b5` on HX370 WSL CPU. One +seed-42 Ryan utterance yielded 117 frames of 16 codebook indices and a +224,640-sample waveform. It decoded frames 0–96 as 72 history plus 25 new +frames through the FP32 decoder and the revision-pinned ONNX export. + +The [local report](report.json) shows ONNX Runtime CPU versus the exact +25-frame eager window at **1.02e-6 relative L2**. The eager window versus +the same segment from an FP32 full 117-frame decode differed by **7.26e-8 +relative L2**. The [code tensor](codes.npz), [quantized fixture](fixture.npz), +[eager](eager_output.npz), [ONNX CPU](ort_output.npz) and +[full-decoder segment](full_reference_segment.npz) are retained. This checks +the window boundary on one real generated-code stream; it is not a speech +quality or mobile pipeline pass. + +The quantized fixture was uploaded as [Workbench dataset](dataset_submission.json) +and [GPU-requested inference](inference_report.json) succeeded against the +exact pinned S25 TFLite target. The [device waveform](device_output.npz) was +finite and unsaturated; the [audit](audit_report.json) measures **1.153% +relative L2 / 38.76 dB SNR** versus local ONNX CPU on this generated-code +window. The separate 100-sample component profile used Workbench-generated +input of the same shape and does not measure this specific fixture. One +generated window does not establish speech intelligibility, all utterances, +or device-local talker/predictor/vocoder integration. diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/audit_report.json b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/audit_report.json new file mode 100644 index 000000000..264a6cbd9 --- /dev/null +++ b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/audit_report.json @@ -0,0 +1,85 @@ +{ + "scope": "one generated-code 25-frame vocoder window on hosted S25; no device-local talker or complete stream", + "status": "component_numeric_measured_no_speech_quality_gate", + "device": { + "name": "Samsung Galaxy S25", + "os": "15", + "attributes": [ + "os:android", + "framework:tflite", + "framework:onnx", + "abi:aarch64-android", + "vendor:samsung", + "format:phone", + "chipset:qualcomm-snapdragon-8-elite-for-galaxy", + "chipset:sm8750-ac", + "hexagon:v79", + "soc-model:69", + "htp-supports-fp16:true", + "htp-supports-weight-sharing:true", + "framework:qnn" + ] + }, + "source_model_id": "mng2vkdrm", + "target_model_id": "mn4o5j17q", + "input_dataset_id": "d2q4y5jl7", + "jobs": { + "compile": "jgddd1zzg", + "inference": "j5688yryg" + }, + "files": { + "fixture": { + "sha256": "4220eec4ee00e5fbb0df6f195d883c6d4549accf399853c3a8e2f83821724452", + "bytes": 185626 + }, + "device": { + "sha256": "7ecafcc289f098aa497f9c9eaa1da0e19879386157e8d2af7c1842e5ea1bfe46", + "bytes": 112200 + }, + "ort": { + "sha256": "df4ad896c62a0d621c9bc48120f929d3561666359a570068b01a3228354411a0", + "bytes": 179186 + }, + "eager": { + "sha256": "8c0c55197daa99433e8a7594525748b4e50e0570ef983dc5b76eedb56936d6d4", + "bytes": 179198 + }, + "full": { + "sha256": "3571033b52c4715761b469e17915d7b1456f5c251322c8a718fad8f905d428e9", + "bytes": 179207 + } + }, + "ort_cpu_vs_device": { + "relative_l2": 0.011533186351546298, + "snr_db": 38.76101381296397, + "max_abs": 0.011052757501602173, + "reference_rms": 0.074840782807901, + "device_rms": 0.07483006310024524, + "device_saturated_fraction_abs_ge_0_999": 0.0 + }, + "eager_cpu_vs_device": { + "relative_l2": 0.011533295679175895, + "snr_db": 38.760931476374985, + "max_abs": 0.011053577065467834, + "reference_rms": 0.0748407755479042, + "device_rms": 0.07483006310024524, + "device_saturated_fraction_abs_ge_0_999": 0.0 + }, + "full_fp32_decoder_segment_vs_device": { + "relative_l2": 0.011533293686009369, + "snr_db": 38.76093297745728, + "max_abs": 0.011053577065467834, + "reference_rms": 0.07484077531515666, + "device_rms": 0.07483006310024524, + "device_saturated_fraction_abs_ge_0_999": 0.0 + }, + "gross_numeric_failure_screen": { + "failed": false, + "criteria": "relative L2 >= 1.0 or >=10% saturated samples; not a speech-quality tolerance" + }, + "limits": [ + "One generated-code window is not a speech listening or intelligibility test.", + "Talker, predictor, co-resident handoff, state, cancellation and memory admission did not run on S25.", + "The separate placement profile used Workbench-generated same-shape input, not this fixture." + ] +} diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/codes.npz b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/codes.npz new file mode 100644 index 000000000..bae5b2f58 Binary files /dev/null and b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/codes.npz differ diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/dataset_submission.json b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/dataset_submission.json new file mode 100644 index 000000000..e380d2938 --- /dev/null +++ b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/dataset_submission.json @@ -0,0 +1,14 @@ +{ + "timestamp_utc": "2026-09-24T06:46:09.898281+00:00", + "facility": "Qualcomm AI Hub Workbench", + "qai_hub_version": "0.55.0", + "fixture_sha256": "4220eec4ee00e5fbb0df6f195d883c6d4549accf399853c3a8e2f83821724452", + "dataset_id": "d2q4y5jl7", + "input_shape": [ + 1, + 512, + 97 + ], + "sample_count": 1, + "status": "UPLOADED" +} diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/device_output.npz b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/device_output.npz new file mode 100644 index 000000000..3d62a3ff9 Binary files /dev/null and b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/device_output.npz differ diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/eager_output.npz b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/eager_output.npz new file mode 100644 index 000000000..38a0be060 Binary files /dev/null and b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/eager_output.npz differ diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/fixture.npz b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/fixture.npz new file mode 100644 index 000000000..e890d18e5 Binary files /dev/null and b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/fixture.npz differ diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/full_reference_segment.npz b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/full_reference_segment.npz new file mode 100644 index 000000000..b090164d5 Binary files /dev/null and b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/full_reference_segment.npz differ diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/inference_report.json b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/inference_report.json new file mode 100644 index 000000000..0b7ffcb08 --- /dev/null +++ b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/inference_report.json @@ -0,0 +1,33 @@ +{ + "timestamp_utc": "2026-09-24T06:53:00.381461+00:00", + "qai_hub_version": "0.55.0", + "facility": "Qualcomm AI Hub Workbench", + "job_id": "j5688yryg", + "url": "https://workbench.aihub.qualcomm.com/jobs/j5688yryg/", + "status": "SUCCESS", + "status_message": "", + "options": "--compute_unit gpu", + "hub_version": "aihub-2026.09.11.0", + "device": { + "name": "Samsung Galaxy S25", + "os": "15", + "attributes": [ + "os:android", + "framework:tflite", + "framework:onnx", + "abi:aarch64-android", + "vendor:samsung", + "format:phone", + "chipset:qualcomm-snapdragon-8-elite-for-galaxy", + "chipset:sm8750-ac", + "hexagon:v79", + "soc-model:69", + "htp-supports-fp16:true", + "htp-supports-weight-sharing:true", + "framework:qnn" + ] + }, + "model_id": "mn4o5j17q", + "input_dataset_id": "d2q4y5jl7", + "outputs": "device_output.npz" +} diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/inference_submission.json b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/inference_submission.json new file mode 100644 index 000000000..5d220e5a5 --- /dev/null +++ b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/inference_submission.json @@ -0,0 +1,12 @@ +{ + "timestamp_utc": "2026-09-24T06:46:12.071383+00:00", + "facility": "Qualcomm AI Hub Workbench", + "qai_hub_version": "0.55.0", + "job_id": "j5688yryg", + "url": "https://workbench.aihub.qualcomm.com/jobs/j5688yryg/", + "model_id": "mn4o5j17q", + "input_dataset_id": "d2q4y5jl7", + "options": "--compute_unit gpu", + "device": "Samsung Galaxy S25", + "status": "SUBMITTED" +} diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/ort_output.npz b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/ort_output.npz new file mode 100644 index 000000000..497707681 Binary files /dev/null and b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/ort_output.npz differ diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/report.json b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/report.json new file mode 100644 index 000000000..53a332b97 --- /dev/null +++ b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/report.json @@ -0,0 +1,74 @@ +{ + "scope": "one real generated-code 25-frame Code2Wav window on local CPU; no device execution", + "model_revision": "85e237c12c027371202489a0ec509ded67b5e4b5", + "model_weight_sha256": "bc3c7e785eb961179c25450d1acff03f839e0002f2f3a5aeb67b5735c0fa2adb", + "decoder_weight_sha256": "836b7b357f5ea43e889936a3709af68dfe3751881acefe4ecf0dbd30ba571258", + "onnx_sha256": "01ee1cd02e5cd264200e5c50767a1805a6180378b051ff9b7ec0d8b6c75d7989", + "torch": "2.13.0+cpu", + "onnxruntime": "1.29.0", + "text": "The small red square is centered on the white card. I can see its straight edges clearly, and the surrounding background is bright and uncluttered.", + "speaker": "Ryan", + "language": "English", + "seed": 42, + "generator_dtype": "bfloat16", + "decoder_dtype": "float32", + "threads": 8, + "generated_frames": 117, + "generated_audio_samples": 224640, + "window": { + "start_frame": 0, + "history_frames": 72, + "new_frames": 25 + }, + "files": { + "codes": { + "filename": "codes.npz", + "sha256": "2768b9274748d767deca67ebd4c0201b914f4839902b89d5693d39e9b8e8c946" + }, + "fixture": { + "filename": "fixture.npz", + "sha256": "4220eec4ee00e5fbb0df6f195d883c6d4549accf399853c3a8e2f83821724452" + }, + "eager": { + "filename": "eager_output.npz", + "sha256": "8c0c55197daa99433e8a7594525748b4e50e0570ef983dc5b76eedb56936d6d4" + }, + "ort": { + "filename": "ort_output.npz", + "sha256": "df4ad896c62a0d621c9bc48120f929d3561666359a570068b01a3228354411a0" + }, + "full_reference": { + "filename": "full_reference_segment.npz", + "sha256": "3571033b52c4715761b469e17915d7b1456f5c251322c8a718fad8f905d428e9" + } + }, + "ort_vs_window_eager": { + "relative_l2": 1.01682301490854e-06, + "snr_db": 119.85509264926462, + "max_abs": 1.1324882507324219e-06, + "reference_rms": 0.0748407755479042, + "observed_rms": 0.074840782807901 + }, + "window_eager_vs_full_fp32_decoder_segment": { + "relative_l2": 7.264658178733583e-08, + "snr_db": 142.7756963116475, + "max_abs": 1.695007085800171e-07, + "reference_rms": 0.07484077531515666, + "observed_rms": 0.0748407755479042 + }, + "limits": [ + "One generated stream and one window do not establish speech quality.", + "The model generates codes on local CPU; this does not test a mobile talker or handoff.", + "The FP32 full decoder reference is separate from the BF16 generator's waveform." + ], + "qwen_tts_sources": { + "wrapper": { + "path": "/home/zhout/project/edge_infer/scratch/qwentts_wsl_pins/qwen_tts/inference/qwen3_tts_model.py", + "sha256": "e1da450732857c1f5fe3e36ebab85db2f6dc6a48caaff6973f463384e30275e4" + }, + "decoder": { + "path": "/home/zhout/project/edge_infer/scratch/qwentts_wsl_pins/qwen_tts/core/tokenizer_12hz/modeling_qwen3_tts_tokenizer_v2.py", + "sha256": "844e8dd8c0182ef9c6463c874631c22ef3c5a4fd1899dd657016164cc5379628" + } + } +} diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/xelite/qnn_short_chunk/README.md b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/xelite/qnn_short_chunk/README.md index 67639bfd6..da594c645 100644 --- a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/xelite/qnn_short_chunk/README.md +++ b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/xelite/qnn_short_chunk/README.md @@ -6,9 +6,14 @@ compiled to [FP16 QNN DLC](compile_report.json) for Snapdragon X Elite CRD. fixture. The [audit](audit_report.json) measures **0.930% relative L2** and **40.63 dB SNR** against local ONNX Runtime CPU, with no saturated samples. This is one synthetic, two-frame waveform without a listening-quality gate. -A separate [placement profile](profile_submission.json) remains in progress -for the same target and tensor shape; the request alone does not verify NPU -node placement or latency. The earlier +A separate [placement profile](profile_report.json) returned 100 component +samples at nearest-rank p50/p95 **3.662/3.676 s** for **0.16 s** audio, +with all 728 execution-detail rows on NPU and 15,855,616 bytes reported +inference peak. Its median component RTF is **22.89**. The profile used +Workbench-generated same-shape input, while the waveform inference used the +retained fixture; the [audit](audit_report.json) binds both jobs to the exact +target. This measured latency is a playable-stream blocker for this artifact. +The earlier [25-frame X Elite NPU attempt](../qnn_npu/inference_report.json) returned a waveform but its placement profile timed out during graph preparation; this shorter artifact tests a separate memory/graph-preparation boundary. Neither diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/xelite/qnn_short_chunk/audit_report.json b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/xelite/qnn_short_chunk/audit_report.json index 54f456fe4..22f990a6c 100644 --- a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/xelite/qnn_short_chunk/audit_report.json +++ b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/xelite/qnn_short_chunk/audit_report.json @@ -1,6 +1,6 @@ { "scope": "one 2-frame Qwen3-TTS vocoder component on hosted Snapdragon X Elite CRD; no complete stream", - "status": "component_inference_numeric_measured", + "status": "component_inference_numeric_and_placement_measured", "device": { "name": "Snapdragon X Elite CRD", "os": "11", @@ -24,7 +24,8 @@ "input_dataset_id": "d7ovqp0w2", "jobs": { "compile": "j5m0m22qg", - "inference": "jgdd3lxeg" + "inference": "jgdd3lxeg", + "profile": "jp4yyrrvp" }, "route": { "compile_options": "--target_runtime qnn_dlc --qnn_options default_graph_htp_precision=FLOAT16", @@ -71,7 +72,17 @@ "limits": [ "One fixed-window waveform has no listening-quality tolerance.", "The 72-frame input history, talker, predictor and stage handoff did not run on the device.", - "No persistent state, admission, cancellation or complete TTS stream was tested.", - "Requested accelerator execution is not verified node placement without a profile." - ] + "No persistent state, admission, cancellation or complete TTS stream was tested." + ], + "profile": { + "sample_count": 100, + "min_us": 3646540, + "nearest_rank_p50_us": 3661691, + "nearest_rank_p95_us": 3676170, + "reported_peak_memory_bytes": 15855616, + "compute_unit_row_counts": { + "NPU": 728 + }, + "profile_input": "Workbench generated input with audited tensor shape; the waveform inference used the retained fixture" + } } diff --git a/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/xelite/qnn_short_chunk/profile_report.json b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/xelite/qnn_short_chunk/profile_report.json new file mode 100644 index 000000000..c88dacc6f --- /dev/null +++ b/benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/xelite/qnn_short_chunk/profile_report.json @@ -0,0 +1,5291 @@ +{ + "timestamp_utc": "2026-09-24T06:51:54.713529+00:00", + "qai_hub_version": "0.55.0", + "facility": "Qualcomm AI Hub Workbench", + "job_id": "jp4yyrrvp", + "url": "https://workbench.aihub.qualcomm.com/jobs/jp4yyrrvp/", + "status": "SUCCESS", + "status_message": "", + "options": "--compute_unit npu", + "hub_version": "aihub-2026.09.11.0", + "device": { + "name": "Snapdragon X Elite CRD", + "os": "11", + "attributes": [ + "os:windows", + "format:compute", + "framework:onnx", + "vendor:qualcomm", + "abi:aarch64-windows", + "chipset:qualcomm-snapdragon-x-elite", + "chipset:sc8380xp", + "hexagon:v73", + "soc-model:60", + "htp-supports-fp16:true", + "htp-supports-weight-sharing:true", + "framework:qnn" + ] + }, + "model_id": "mqeyv2w7m", + "shapes": { + "quantized": [ + [ + 1, + 512, + 74 + ], + "float32" + ] + }, + "profile": { + "execution_summary": { + "estimated_inference_time": 3646540, + "estimated_inference_peak_memory": 15855616, + "first_load_time": 787624177, + "first_load_peak_memory": 7648456704, + "warm_load_time": 1883699, + "warm_load_peak_memory": 15900672, + "compile_time": 0, + "compile_peak_memory": 0, + "compile_memory_increase_range": [ + 0, + 0 + ], + "compile_memory_peak_range": [ + 0, + 0 + ], + "first_load_memory_increase_range": [ + 6978572288, + 6978572288 + ], + "first_load_memory_peak_range": [ + 7643504640, + 7643504640 + ], + "warm_load_memory_increase_range": [ + 9932800, + 9932800 + ], + "warm_load_memory_peak_range": [ + 10932224, + 10932224 + ], + "inference_memory_increase_range": [ + 0, + 0 + ], + "inference_memory_peak_range": [ + 806912, + 806912 + ], + "all_compile_times": [], + "all_first_load_times": [ + 787624177 + ], + "all_warm_load_times": [ + 1883699 + ], + "all_inference_times": [ + 3659081, + 3666666, + 3664154, + 3668243, + 3661363, + 3661931, + 3666330, + 3663360, + 3661124, + 3654547, + 3659404, + 3662460, + 3656818, + 3666579, + 3670520, + 3660665, + 3664483, + 3660183, + 3665463, + 3663760, + 3674075, + 3655929, + 3663775, + 3656077, + 3659013, + 3656390, + 3657209, + 3653841, + 3657847, + 3663713, + 3654056, + 3662673, + 3683953, + 3661735, + 3662218, + 3655867, + 3664092, + 3661314, + 3671687, + 3663565, + 3654582, + 3661584, + 3662412, + 3665981, + 3660089, + 3664790, + 3666849, + 3676170, + 3657022, + 3665250, + 3657358, + 3657743, + 3660100, + 3660207, + 3665734, + 3653826, + 3655193, + 3665795, + 3646540, + 3657929, + 3661725, + 3654083, + 3666116, + 3674557, + 3661995, + 3660632, + 3653063, + 3660364, + 3667839, + 3664223, + 3651541, + 3664601, + 3661374, + 3656282, + 3692393, + 3655454, + 3655418, + 3654150, + 3654138, + 3669138, + 3651928, + 3666233, + 3664950, + 3672136, + 3658234, + 3689932, + 3674445, + 3660741, + 3681250, + 3656746, + 3671298, + 3655112, + 3672092, + 3661691, + 3656607, + 3678099, + 3658125, + 3652597, + 3668962, + 3663641 + ] + }, + "execution_detail": [ + { + "name": "Input", + "type": "Input", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 7921 + }, + { + "name": "node_conv1d_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 41957 + }, + { + "name": "node_conv1d_reshape_to_2d_0231", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 89044 + }, + { + "name": "node_conv1d_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 346570 + }, + { + "name": "node_conv1d_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_3", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 58019 + }, + { + "name": "node_MatMul_3_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "rms_norm_node___809316", + "type": "rms", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 44391 + }, + { + "name": "node_MatMul_177_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_177", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4145 + }, + { + "name": "node_MatMul_177_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_1_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 57962 + }, + { + "name": "node_Split_1_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 55103 + }, + { + "name": "node_Split_1_to_slice_2", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 94445 + }, + { + "name": "node_view", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_2", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_3", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_2", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_4", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_4", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 66756 + }, + { + "name": "node_Split_139_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_139_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_neg", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 12020 + }, + { + "name": "node_cat_3", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_5", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 52144 + }, + { + "name": "node_add_5", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 2237 + }, + { + "name": "node_mul_6", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 119694 + }, + { + "name": "node_Split_141_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_141_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_neg_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 25275 + }, + { + "name": "node_cat_4", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_7", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 78089 + }, + { + "name": "node_add_6", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 16034 + }, + { + "name": "node_Transpose_105", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_66", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 170407 + }, + { + "name": "node_Mul_68", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 80867 + }, + { + "name": "node_Add_236", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 104768 + }, + { + "name": "node_Softmax_237", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 168998 + }, + { + "name": "node_scaled_dot_product_attention", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 69386 + }, + { + "name": "node_transpose_5", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_3", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_4", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 22269 + }, + { + "name": "node_linear_4_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_8", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 87105 + }, + { + "name": "node_add_7", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 14340 + }, + { + "name": "rms_norm_node_", + "type": "rms", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 23412 + }, + { + "name": "node_linear_6_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_5_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_5", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4262 + }, + { + "name": "node_linear_5_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Sigmoid_251", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_silu", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 47101 + }, + { + "name": "node_linear_6", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 37084 + }, + { + "name": "node_linear_6_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_11", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 23169 + }, + { + "name": "node_linear_7_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_7", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 20100 + }, + { + "name": "node_linear_7_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_12", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 53300 + }, + { + "name": "node_add_9", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 12800 + }, + { + "name": "rms_norm_node___909696", + "type": "rms", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 21450 + }, + { + "name": "node_MatMul_181_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_181", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4198 + }, + { + "name": "node_MatMul_181_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_3_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 66635 + }, + { + "name": "node_Split_3_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 61730 + }, + { + "name": "node_Split_3_to_slice_2", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 81675 + }, + { + "name": "node_view_4", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_6", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_5", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_7", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_6", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_8", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_15", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 66758 + }, + { + "name": "node_Split_143_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_143_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_neg_2", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 12559 + }, + { + "name": "node_cat_5", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_16", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 52702 + }, + { + "name": "node_add_11", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 2296 + }, + { + "name": "node_mul_17", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 75121 + }, + { + "name": "node_Split_145_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_145_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_neg_3", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 13259 + }, + { + "name": "node_cat_6", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_18", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 58070 + }, + { + "name": "node_add_12", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 11259 + }, + { + "name": "node_Transpose_109", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_69", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 159434 + }, + { + "name": "node_Mul_71", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 71245 + }, + { + "name": "node_Add_340", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 90361 + }, + { + "name": "node_Softmax_341", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 134387 + }, + { + "name": "node_scaled_dot_product_attention_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 58591 + }, + { + "name": "node_transpose_9", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_7", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_11", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 20931 + }, + { + "name": "node_linear_11_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_19", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 69731 + }, + { + "name": "node_add_13", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 14830 + }, + { + "name": "rms_norm_node___236703", + "type": "rms", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 23430 + }, + { + "name": "node_linear_13_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_12_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_12", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4270 + }, + { + "name": "node_linear_12_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Sigmoid_355", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_silu_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 45432 + }, + { + "name": "node_linear_13", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 37913 + }, + { + "name": "node_linear_13_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_22", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 9525 + }, + { + "name": "node_linear_14_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_14", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 18790 + }, + { + "name": "node_linear_14_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_23", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 59162 + }, + { + "name": "node_add_15", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 12560 + }, + { + "name": "rms_norm_node___612561", + "type": "rms", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 21103 + }, + { + "name": "node_MatMul_185_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_185", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4246 + }, + { + "name": "node_MatMul_185_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_5_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 65859 + }, + { + "name": "node_Split_5_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 60565 + }, + { + "name": "node_Split_5_to_slice_2", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 80043 + }, + { + "name": "node_view_8", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_10", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_9", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_11", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_10", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_12", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_26", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 67632 + }, + { + "name": "node_Split_147_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_147_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_neg_4", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 12321 + }, + { + "name": "node_cat_7", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_27", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 53352 + }, + { + "name": "node_add_17", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 2235 + }, + { + "name": "node_mul_28", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 75212 + }, + { + "name": "node_Split_149_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_149_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_neg_5", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 14215 + }, + { + "name": "node_cat_8", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_29", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 58943 + }, + { + "name": "node_add_18", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 11098 + }, + { + "name": "node_Transpose_113", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_72", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 157254 + }, + { + "name": "node_Mul_74", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 66652 + }, + { + "name": "node_Add_444", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 135798 + }, + { + "name": "node_Softmax_445", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 135065 + }, + { + "name": "node_scaled_dot_product_attention_2", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 58094 + }, + { + "name": "node_transpose_13", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_11", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_18", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 23318 + }, + { + "name": "node_linear_18_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_30", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 77677 + }, + { + "name": "node_add_19", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 13906 + }, + { + "name": "rms_norm_node___608555", + "type": "rms", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 21528 + }, + { + "name": "node_linear_20_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_19_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_19", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4205 + }, + { + "name": "node_linear_19_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Sigmoid_459", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_silu_2", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 46225 + }, + { + "name": "node_linear_20", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 37391 + }, + { + "name": "node_linear_20_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_33", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 9638 + }, + { + "name": "node_linear_21_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_21", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 18732 + }, + { + "name": "node_linear_21_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_34", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 53947 + }, + { + "name": "node_add_21", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 12605 + }, + { + "name": "rms_norm_node___319690", + "type": "rms", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 21151 + }, + { + "name": "node_MatMul_189_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_189", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4173 + }, + { + "name": "node_MatMul_189_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_7_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 60971 + }, + { + "name": "node_Split_7_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 61834 + }, + { + "name": "node_Split_7_to_slice_2", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 79579 + }, + { + "name": "node_view_12", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_14", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_13", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_15", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_14", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_16", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_37", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 66960 + }, + { + "name": "node_Split_151_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_151_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_neg_6", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 13002 + }, + { + "name": "node_cat_9", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_38", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 52595 + }, + { + "name": "node_add_23", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 2172 + }, + { + "name": "node_mul_39", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 74581 + }, + { + "name": "node_Split_153_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_153_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_neg_7", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 13528 + }, + { + "name": "node_cat_10", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_40", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 57266 + }, + { + "name": "node_add_24", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 11523 + }, + { + "name": "node_Transpose_117", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_75", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 156623 + }, + { + "name": "node_Mul_77", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 66231 + }, + { + "name": "node_Add_548", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 89912 + }, + { + "name": "node_Softmax_549", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 134631 + }, + { + "name": "node_scaled_dot_product_attention_3", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 55978 + }, + { + "name": "node_transpose_17", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_15", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_25", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 21271 + }, + { + "name": "node_linear_25_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_41", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 66558 + }, + { + "name": "node_add_25", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 13220 + }, + { + "name": "rms_norm_node___915689", + "type": "rms", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 21456 + }, + { + "name": "node_linear_27_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_26_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_26", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4235 + }, + { + "name": "node_linear_26_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Sigmoid_563", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_silu_3", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 46035 + }, + { + "name": "node_linear_27", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 37196 + }, + { + "name": "node_linear_27_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_44", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 9896 + }, + { + "name": "node_linear_28_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_28", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 19147 + }, + { + "name": "node_linear_28_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_45", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 57857 + }, + { + "name": "node_add_27", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 13606 + }, + { + "name": "rms_norm_node___738921", + "type": "rms", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 21070 + }, + { + "name": "node_MatMul_193_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_193", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4647 + }, + { + "name": "node_MatMul_193_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_9_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 62818 + }, + { + "name": "node_Split_9_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 58903 + }, + { + "name": "node_Split_9_to_slice_2", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 80269 + }, + { + "name": "node_view_16", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_18", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_17", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_19", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_18", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_20", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_48", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 66801 + }, + { + "name": "node_Split_155_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_155_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_neg_8", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 13360 + }, + { + "name": "node_cat_11", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_49", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 53302 + }, + { + "name": "node_add_29", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 2465 + }, + { + "name": "node_mul_50", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 75975 + }, + { + "name": "node_Split_157_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_157_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_neg_9", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 14673 + }, + { + "name": "node_cat_12", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_51", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 59290 + }, + { + "name": "node_add_30", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 10855 + }, + { + "name": "node_Transpose_121", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_78", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 158455 + }, + { + "name": "node_Mul_80", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 70005 + }, + { + "name": "node_Add_652", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 91290 + }, + { + "name": "node_Softmax_653", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 134384 + }, + { + "name": "node_scaled_dot_product_attention_4", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 60307 + }, + { + "name": "node_transpose_21", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_19", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_32", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 22966 + }, + { + "name": "node_linear_32_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_52", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 64535 + }, + { + "name": "node_add_31", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 13415 + }, + { + "name": "rms_norm_node___801484", + "type": "rms", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 21577 + }, + { + "name": "node_linear_34_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_33_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_33", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4298 + }, + { + "name": "node_linear_33_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Sigmoid_667", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_silu_4", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 46354 + }, + { + "name": "node_linear_34", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 37990 + }, + { + "name": "node_linear_34_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_55", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 9705 + }, + { + "name": "node_linear_35_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_35", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 20897 + }, + { + "name": "node_linear_35_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_56", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 49787 + }, + { + "name": "node_add_33", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 13200 + }, + { + "name": "rms_norm_node___106532", + "type": "rms", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 21360 + }, + { + "name": "node_MatMul_197_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_197", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4905 + }, + { + "name": "node_MatMul_197_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_11_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 63579 + }, + { + "name": "node_Split_11_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 58242 + }, + { + "name": "node_Split_11_to_slice_2", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 81358 + }, + { + "name": "node_view_20", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_22", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_21", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_23", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_22", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_24", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_59", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 66767 + }, + { + "name": "node_Split_159_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_159_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_neg_10", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 13608 + }, + { + "name": "node_cat_13", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_60", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 53789 + }, + { + "name": "node_add_35", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 2179 + }, + { + "name": "node_mul_61", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 72078 + }, + { + "name": "node_Split_161_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_161_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_neg_11", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 14610 + }, + { + "name": "node_cat_14", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_62", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 58146 + }, + { + "name": "node_add_36", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 10775 + }, + { + "name": "node_Transpose_125", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_81", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 161926 + }, + { + "name": "node_Mul_83", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 69527 + }, + { + "name": "node_Add_756", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 90152 + }, + { + "name": "node_Softmax_757", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 134058 + }, + { + "name": "node_scaled_dot_product_attention_5", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 57216 + }, + { + "name": "node_transpose_25", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_23", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_39", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 20920 + }, + { + "name": "node_linear_39_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_63", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 59079 + }, + { + "name": "node_add_37", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 12787 + }, + { + "name": "rms_norm_node___651882", + "type": "rms", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 23974 + }, + { + "name": "node_linear_41_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_40_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_40", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4586 + }, + { + "name": "node_linear_40_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Sigmoid_771", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_silu_5", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 46431 + }, + { + "name": "node_linear_41", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 38220 + }, + { + "name": "node_linear_41_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_66", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 9531 + }, + { + "name": "node_linear_42_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_42", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 18749 + }, + { + "name": "node_linear_42_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_67", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 57010 + }, + { + "name": "node_add_39", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 13038 + }, + { + "name": "rms_norm_node___789068", + "type": "rms", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 21348 + }, + { + "name": "node_MatMul_201_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_201", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4532 + }, + { + "name": "node_MatMul_201_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_13_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 57516 + }, + { + "name": "node_Split_13_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 56797 + }, + { + "name": "node_Split_13_to_slice_2", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 80181 + }, + { + "name": "node_view_24", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_26", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_25", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_27", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_26", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_28", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_70", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 68014 + }, + { + "name": "node_Split_163_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_163_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_neg_12", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 14360 + }, + { + "name": "node_cat_15", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_71", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 54497 + }, + { + "name": "node_add_41", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 2214 + }, + { + "name": "node_mul_72", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 71879 + }, + { + "name": "node_Split_165_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_165_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_neg_13", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 14047 + }, + { + "name": "node_cat_16", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_73", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 55498 + }, + { + "name": "node_add_42", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 11437 + }, + { + "name": "node_Transpose_129", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_84", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 161172 + }, + { + "name": "node_Mul_86", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 68495 + }, + { + "name": "node_Add_860", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 90201 + }, + { + "name": "node_Softmax_861", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 134636 + }, + { + "name": "node_scaled_dot_product_attention_6", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 57556 + }, + { + "name": "node_transpose_29", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_27", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_46", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 20113 + }, + { + "name": "node_linear_46_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_74", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 56714 + }, + { + "name": "node_add_43", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 14091 + }, + { + "name": "rms_norm_node___480049", + "type": "rms", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 21871 + }, + { + "name": "node_linear_48_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_47_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_47", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 5068 + }, + { + "name": "node_linear_47_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Sigmoid_875", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_silu_6", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 47566 + }, + { + "name": "node_linear_48", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 37459 + }, + { + "name": "node_linear_48_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_77", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 9989 + }, + { + "name": "node_linear_49_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_49", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 19502 + }, + { + "name": "node_linear_49_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_78", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 52634 + }, + { + "name": "node_add_45", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 14689 + }, + { + "name": "rms_norm_node___889670", + "type": "rms", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 21249 + }, + { + "name": "node_MatMul_205_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_205", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4548 + }, + { + "name": "node_MatMul_205_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_15_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 64478 + }, + { + "name": "node_Split_15_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 60819 + }, + { + "name": "node_Split_15_to_slice_2", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 79609 + }, + { + "name": "node_view_28", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_30", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_29", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_31", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_30", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_transpose_32", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_81", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 66564 + }, + { + "name": "node_Split_167_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_167_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_neg_14", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 14214 + }, + { + "name": "node_cat_17", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_82", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 56378 + }, + { + "name": "node_add_47", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 2249 + }, + { + "name": "node_mul_83", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 70851 + }, + { + "name": "node_Split_169_to_slice_0", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Split_169_to_slice_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_neg_15", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 13585 + }, + { + "name": "node_cat_18", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_84", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 60431 + }, + { + "name": "node_add_48", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 11319 + }, + { + "name": "node_Transpose_133", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_87", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 163313 + }, + { + "name": "node_Mul_89", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 77551 + }, + { + "name": "node_Add_964", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 90812 + }, + { + "name": "node_Softmax_965", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 134526 + }, + { + "name": "node_scaled_dot_product_attention_7", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 58492 + }, + { + "name": "node_transpose_33", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_31", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_53", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 21176 + }, + { + "name": "node_linear_53_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_85", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 75367 + }, + { + "name": "node_add_49", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 14847 + }, + { + "name": "rms_norm_node___409462", + "type": "rms", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 21527 + }, + { + "name": "node_linear_55_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_54_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_54", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4658 + }, + { + "name": "node_linear_54_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Sigmoid_979", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_silu_7", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 46774 + }, + { + "name": "node_linear_55", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 45799 + }, + { + "name": "node_linear_55_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_88", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 13193 + }, + { + "name": "node_linear_56_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_linear_56", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 21193 + }, + { + "name": "node_linear_56_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_89", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 74595 + }, + { + "name": "node_add_51", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 12997 + }, + { + "name": "rms_norm_node___485919", + "type": "rms", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 24054 + }, + { + "name": "node_MatMul_65_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_65", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 49397 + }, + { + "name": "node_MatMul_65_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_convolution_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 174049 + }, + { + "name": "node_convolution_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_1_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_1_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 78054 + }, + { + "name": "node_conv1d_1_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_layer_norm", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 233498 + }, + { + "name": "node_MatMul_991_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_991", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 15850 + }, + { + "name": "node_MatMul_991_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Gelu_172", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 267465 + }, + { + "name": "node_MatMul_207_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_207", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 394031 + }, + { + "name": "node_MatMul_207_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_add_53", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 38509 + }, + { + "name": "node_convolution_1_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_convolution_1_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 19753757 + }, + { + "name": "node_convolution_1_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_2_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_2_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 265478 + }, + { + "name": "node_conv1d_2_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_layer_norm_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 402546 + }, + { + "name": "node_MatMul_1004_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_1004", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_1004_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Gelu_173", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 500300 + }, + { + "name": "node_MatMul_209_pre_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_MatMul_209", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 611606 + }, + { + "name": "node_MatMul_209_post_reshape", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_add_54", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 225937 + }, + { + "name": "node_conv1d_3_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_3_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 8888576 + }, + { + "name": "node_conv1d_3_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_94", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 1828650 + }, + { + "name": "node_sin_1", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 2152663 + }, + { + "name": "node_Mul_33", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 323372 + }, + { + "name": "node_mul_95", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 1990665 + }, + { + "name": "node_add_55", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 523096 + }, + { + "name": "node_convolution_2_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_convolution_2_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 37197411 + }, + { + "name": "node_convolution_2_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_slice_39", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 751651 + }, + { + "name": "node_mul_96", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 7036931 + }, + { + "name": "node_sin_2", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 8844173 + }, + { + "name": "node_Mul_34", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 954638 + }, + { + "name": "node_mul_97", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 7605436 + }, + { + "name": "node_add_56", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 2163860 + }, + { + "name": "node_conv1d_4_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_4_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 28453274 + }, + { + "name": "node_conv1d_4_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_98", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 7367540 + }, + { + "name": "node_sin_3", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 9445152 + }, + { + "name": "node_Mul_35", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 901082 + }, + { + "name": "node_mul_99", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 7766742 + }, + { + "name": "node_add_57", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 44117 + }, + { + "name": "node_conv1d_5_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_5_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 463113 + }, + { + "name": "node_conv1d_5_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_add_58", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 644416 + }, + { + "name": "node_mul_100", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 5667178 + }, + { + "name": "node_sin_4", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 8453129 + }, + { + "name": "node_Mul_36", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 1273134 + }, + { + "name": "node_mul_101", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 7980988 + }, + { + "name": "node_add_59", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 1996770 + }, + { + "name": "node_conv1d_6_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_6_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 30786329 + }, + { + "name": "node_conv1d_6_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_102", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 7070056 + }, + { + "name": "node_sin_5", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 8885265 + }, + { + "name": "node_Mul_37", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 914913 + }, + { + "name": "node_mul_103", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 7777867 + }, + { + "name": "node_add_60", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 69185 + }, + { + "name": "node_conv1d_7_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_7_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 455587 + }, + { + "name": "node_conv1d_7_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_add_61", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 662596 + }, + { + "name": "node_mul_104", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 5960193 + }, + { + "name": "node_sin_6", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 8608138 + }, + { + "name": "node_Mul_38", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 1533921 + }, + { + "name": "node_mul_105", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 8174935 + }, + { + "name": "node_add_62", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 2032102 + }, + { + "name": "node_conv1d_8_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_8_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 33248177 + }, + { + "name": "node_conv1d_8_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_106", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 7073120 + }, + { + "name": "node_sin_7", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 8800456 + }, + { + "name": "node_Mul_39", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 916219 + }, + { + "name": "node_mul_107", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 7791679 + }, + { + "name": "node_add_63", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 45295 + }, + { + "name": "node_conv1d_9_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_9_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 458133 + }, + { + "name": "node_conv1d_9_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_add_64", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 606243 + }, + { + "name": "node_mul_108", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 5765112 + }, + { + "name": "node_sin_8", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 8470232 + }, + { + "name": "node_Mul_40", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 1320517 + }, + { + "name": "node_mul_109", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 7951979 + }, + { + "name": "node_add_65", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 5099015 + }, + { + "name": "node_convolution_3_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_convolution_3_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 34948682 + }, + { + "name": "node_convolution_3_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_slice_40", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 1208760 + }, + { + "name": "node_mul_110", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 19811340 + }, + { + "name": "node_sin_9", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 22181528 + }, + { + "name": "node_Mul_41", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 5894420 + }, + { + "name": "node_mul_111", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 19237010 + }, + { + "name": "node_add_66", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 6138931 + }, + { + "name": "node_conv1d_10_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_10_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 42284156 + }, + { + "name": "node_conv1d_10_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_112", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 18225997 + }, + { + "name": "node_sin_10", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 22472234 + }, + { + "name": "node_Mul_42", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 5292813 + }, + { + "name": "node_mul_113", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 18777622 + }, + { + "name": "node_add_67", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 501562 + }, + { + "name": "node_conv1d_11_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_11_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 843878 + }, + { + "name": "node_conv1d_11_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_add_68", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4851188 + }, + { + "name": "node_mul_114", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 16765555 + }, + { + "name": "node_sin_11", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 21294099 + }, + { + "name": "node_Mul_43", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 5551221 + }, + { + "name": "node_mul_115", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 19221005 + }, + { + "name": "node_add_69", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 6143656 + }, + { + "name": "node_conv1d_12_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_12_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 44825047 + }, + { + "name": "node_conv1d_12_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_116", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 20277025 + }, + { + "name": "node_sin_12", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 21753326 + }, + { + "name": "node_Mul_44", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 5863820 + }, + { + "name": "node_mul_117", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 20039613 + }, + { + "name": "node_add_70", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 498502 + }, + { + "name": "node_conv1d_13_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_13_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 1018349 + }, + { + "name": "node_conv1d_13_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_add_71", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4971153 + }, + { + "name": "node_mul_118", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 17624309 + }, + { + "name": "node_sin_13", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 21436122 + }, + { + "name": "node_Mul_45", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 5531672 + }, + { + "name": "node_mul_119", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 19844133 + }, + { + "name": "node_add_72", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 5812923 + }, + { + "name": "node_conv1d_14_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_14_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 47023942 + }, + { + "name": "node_conv1d_14_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_120", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 20038152 + }, + { + "name": "node_sin_14", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 22008991 + }, + { + "name": "node_Mul_46", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 6342629 + }, + { + "name": "node_mul_121", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 20082183 + }, + { + "name": "node_add_73", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 606528 + }, + { + "name": "node_conv1d_15_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_15_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 1020503 + }, + { + "name": "node_conv1d_15_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_add_74", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4945193 + }, + { + "name": "node_mul_122", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 17664227 + }, + { + "name": "node_sin_15", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 21408385 + }, + { + "name": "node_Mul_47", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 5764467 + }, + { + "name": "node_mul_123", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 19982169 + }, + { + "name": "node_add_75", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 14467782 + }, + { + "name": "node_convolution_4_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_convolution_4_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 50705924 + }, + { + "name": "node_convolution_4_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_slice_41", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 9557446 + }, + { + "name": "node_mul_124", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 36650800 + }, + { + "name": "node_sin_16", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 40992889 + }, + { + "name": "node_Mul_48", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 13015191 + }, + { + "name": "node_mul_125", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 48480018 + }, + { + "name": "node_add_76", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 6836086 + }, + { + "name": "node_conv1d_16_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_16_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 47209763 + }, + { + "name": "node_conv1d_16_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_126", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 36841932 + }, + { + "name": "node_sin_17", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 41456228 + }, + { + "name": "node_Mul_49", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 12932957 + }, + { + "name": "node_mul_127", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 48572946 + }, + { + "name": "node_add_77", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 5293575 + }, + { + "name": "node_conv1d_17_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_17_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 584446 + }, + { + "name": "node_conv1d_17_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_add_78", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4070085 + }, + { + "name": "node_mul_128", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 29470449 + }, + { + "name": "node_sin_18", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 41405496 + }, + { + "name": "node_Mul_50", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 12173142 + }, + { + "name": "node_mul_129", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 48258724 + }, + { + "name": "node_add_79", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4098871 + }, + { + "name": "node_conv1d_18_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_18_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 49998068 + }, + { + "name": "node_conv1d_18_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_130", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 40047637 + }, + { + "name": "node_sin_19", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 45152869 + }, + { + "name": "node_Mul_51", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 10679308 + }, + { + "name": "node_mul_131", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 49107801 + }, + { + "name": "node_add_80", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 5829966 + }, + { + "name": "node_conv1d_19_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_19_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 596304 + }, + { + "name": "node_conv1d_19_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_add_81", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 2972197 + }, + { + "name": "node_mul_132", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 29268261 + }, + { + "name": "node_sin_20", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 41209581 + }, + { + "name": "node_Mul_52", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 12133670 + }, + { + "name": "node_mul_133", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 49200017 + }, + { + "name": "node_add_82", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 4177139 + }, + { + "name": "node_conv1d_20_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_20_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 50500768 + }, + { + "name": "node_conv1d_20_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_134", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 38313370 + }, + { + "name": "node_sin_21", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 45170799 + }, + { + "name": "node_Mul_53", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 10442166 + }, + { + "name": "node_mul_135", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 49148939 + }, + { + "name": "node_add_83", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 5748540 + }, + { + "name": "node_conv1d_21_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_21_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 594227 + }, + { + "name": "node_conv1d_21_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_add_84", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 2475325 + }, + { + "name": "node_mul_136", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 29493494 + }, + { + "name": "node_sin_22", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 40693084 + }, + { + "name": "node_Mul_54", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 11695305 + }, + { + "name": "node_mul_137", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 49168751 + }, + { + "name": "node_add_85", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 32967115 + }, + { + "name": "node_convolution_5_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_convolution_5_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 67364879 + }, + { + "name": "node_convolution_5_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_slice_42", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 267296225 + }, + { + "name": "node_mul_138", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 57430915 + }, + { + "name": "node_sin_23", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 67772384 + }, + { + "name": "node_Mul_55", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 39877145 + }, + { + "name": "node_mul_139", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 390384494 + }, + { + "name": "node_add_86", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 62160569 + }, + { + "name": "node_conv1d_22_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_22_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 304253798 + }, + { + "name": "node_conv1d_22_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_140", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 57948173 + }, + { + "name": "node_sin_24", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 67717837 + }, + { + "name": "node_Mul_56", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 37653005 + }, + { + "name": "node_mul_141", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 453374200 + }, + { + "name": "node_add_87", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 48839266 + }, + { + "name": "node_conv1d_23_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_23_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 1492741 + }, + { + "name": "node_conv1d_23_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_add_88", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 40999892 + }, + { + "name": "node_mul_142", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 97290024 + }, + { + "name": "node_sin_25", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 191978799 + }, + { + "name": "node_Mul_57", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 92739543 + }, + { + "name": "node_mul_143", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 393540535 + }, + { + "name": "node_add_89", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 55708592 + }, + { + "name": "node_conv1d_24_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_24_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 358654639 + }, + { + "name": "node_conv1d_24_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_144", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 52820870 + }, + { + "name": "node_sin_26", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 67932534 + }, + { + "name": "node_Mul_58", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 42190338 + }, + { + "name": "node_mul_145", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 460064765 + }, + { + "name": "node_add_90", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 49535296 + }, + { + "name": "node_conv1d_25_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_25_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 1539043 + }, + { + "name": "node_conv1d_25_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_add_91", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 26698311 + }, + { + "name": "node_mul_146", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 95396375 + }, + { + "name": "node_sin_27", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 196347633 + }, + { + "name": "node_Mul_59", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 97175782 + }, + { + "name": "node_mul_147", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 394091109 + }, + { + "name": "node_add_92", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 54948680 + }, + { + "name": "node_conv1d_26_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_26_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 359911537 + }, + { + "name": "node_conv1d_26_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_mul_148", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 52933210 + }, + { + "name": "node_sin_28", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 67785475 + }, + { + "name": "node_Mul_60", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 41923672 + }, + { + "name": "node_mul_149", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 461075782 + }, + { + "name": "node_add_93", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 49253937 + }, + { + "name": "node_conv1d_27_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_27_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 1516241 + }, + { + "name": "node_conv1d_27_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_add_94", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 25689288 + }, + { + "name": "node_mul_150", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 95371803 + }, + { + "name": "node_sin_29", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 193833438 + }, + { + "name": "node_Mul_61", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 94810038 + }, + { + "name": "node_mul_151", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 394503083 + }, + { + "name": "node_add_95", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 36622528 + }, + { + "name": "node_conv1d_28_reshape_to_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_28_2d", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_conv1d_28_intermediate", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_Clip_1310", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 197864 + }, + { + "name": "clamp_012", + "type": "clamp", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_view_32", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 0 + }, + { + "name": "node_slice_43", + "type": "node", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 61174 + }, + { + "name": "Output", + "type": "Output", + "compute_unit": "NPU", + "execution_time": 0, + "execution_cycles": 5067 + } + ] + } +} diff --git a/benchmarks/edge_harness/results/e2e_profiling_20260922/evidence/summary/README.md b/benchmarks/edge_harness/results/e2e_profiling_20260922/evidence/summary/README.md index 316c5d298..2096b838a 100644 --- a/benchmarks/edge_harness/results/e2e_profiling_20260922/evidence/summary/README.md +++ b/benchmarks/edge_harness/results/e2e_profiling_20260922/evidence/summary/README.md @@ -49,8 +49,8 @@ NOT E2E means no complete device/model workload was verified. “Scoped E2E” i | PC HX370 + Radeon 890M / Windows worker | scoped E2E: Omni Spark text | scoped E2E: Omni Radeon talker/codec + CPU predictor | NOT E2E | scoped E2E: Omni Radeon+CPU image+audio → text+speech (native-storage 20-run profile) | synthetic policy pass: Omni CPU+Radeon Cosmos (WSL/native Windows) | | PC HX370 + AMD NPU / Windows worker | NOT E2E (output-head component passed) | NOT E2E | NOT E2E | NOT E2E | NOT E2E | | PC CPU+iGPU+NPU +/- discrete GPU, joint model execution | NOT E2E | NOT E2E | NOT E2E | NOT E2E | NOT E2E | -| PC Snapdragon X Elite CRD / AI Hub | NOT E2E (one Spark attention layer on NPU) | NOT E2E (CPU talker/vocoder steps; two-frame QNN vocoder waveform measured, placement pending) | NOT E2E | NOT E2E (NPU speech-head step profiled) | NOT E2E (Cosmos NPU encoder profiled) | -| Mobile Galaxy S25 / Snapdragon 8 Elite for Galaxy | NOT E2E (one Spark attention layer on NPU) | NOT E2E (FP16 talker token mismatch; CPU talker + vocoder profiled; 25-frame and two-frame QNN vocoder memory failures; historical 25-frame TFLite mixed vocoder component RTF 0.420; pinned 25-frame GPU waveform measured, placement pending; short chunk RTF 3.49) | NOT E2E | NOT E2E (NPU speech head step profiled) | NOT E2E (Cosmos NPU encoder profiled) | +| PC Snapdragon X Elite CRD / AI Hub | NOT E2E (one Spark attention layer on NPU) | NOT E2E (CPU talker/vocoder steps; two-frame QNN vocoder on NPU, component RTF 22.89) | NOT E2E | NOT E2E (NPU speech-head step profiled) | NOT E2E (Cosmos NPU encoder profiled) | +| Mobile Galaxy S25 / Snapdragon 8 Elite for Galaxy | NOT E2E (one Spark attention layer on NPU) | NOT E2E (FP16 talker token mismatch; CPU talker + vocoder profiled; 25-frame and two-frame QNN vocoder memory failures; historical 25-frame TFLite mixed vocoder component RTF 0.420; pinned 25-frame mixed CPU/GPU vocoder profiled at RTF 0.335; generated-code waveform measured; short chunk RTF 3.49) | NOT E2E | NOT E2E (NPU speech head step profiled) | NOT E2E (Cosmos NPU encoder profiled) | | Mobile Galaxy S24 / Snapdragon 8 Gen 3 | NOT E2E (one Spark attention layer on NPU) | NOT E2E (FP16 talker mismatch; short CPU vocoder RTF 21.46; mixed CPU/GPU waveform saturated) | NOT E2E | NOT E2E (NPU speech head step profiled) | NOT E2E (Cosmos encoder numeric and NPU profiled) | | Embedded SA8775P ADP / AI Hub | NOT E2E (one Spark attention layer on NPU) | NOT E2E (CPU talker/vocoder steps; NPU vocoder failed) | NOT E2E | NOT E2E (NPU speech-head step profiled) | NOT E2E (Cosmos NPU encoder profiled) | | Embedded RB3 Gen 2 / QCS6490 / AI Hub | NOT E2E (Spark CPU layer passed; W8A8 NPU quality open) | NOT E2E (CPU talker + vocoder profiled; INT8 NPU vocoder fails numeric gate) | NOT E2E | NOT E2E (CPU speech-head step profiled) | NOT E2E (Cosmos CPU numeric and profiled) | @@ -95,12 +95,12 @@ NOT E2E means no complete device/model workload was verified. “Scoped E2E” i | PC CPU+iGPU+NPU +/- discrete GPU, joint model execution | MiniCPM-o 4.5 | NOT_E2E_PROFILEABLE | No qualifying same-model multi-accelerator deployment/artifact plan or overlap/recovery evidence. Individual routes are not a joint pipeline. | Supply and validate the missing matching artifact/backend, then run the complete specified workload. | | PC CPU+iGPU+NPU +/- discrete GPU, joint model execution | InternVLA-A1 | NOT_E2E_PROFILEABLE | No qualifying same-model multi-accelerator deployment/artifact plan or overlap/recovery evidence. Individual routes are not a joint pipeline. | Supply and validate the missing matching artifact/backend, then run the complete specified workload. | | PC Snapdragon X Elite CRD / AI Hub | Spark-X2.5 | COMPONENT_NPU_INFERENCE_PROFILED; NOT_E2E_PROFILEABLE | A real Spark-X2.5-1.7B one-layer W8A16 QDQ attention artifact, compiled as QNN DLC, ran on the exact hosted Snapdragon X Elite CRD (Windows 11 ARM64). Inference returned finite hidden/new-K/new-V tensors bitwise identical to the prior S24/S25 NPU outputs on the same fixture; the profile attributed all 73 execution-detail rows to NPU. Across 100 component samples, nearest-rank p50/p95 was 1.433/1.606 ms; the 1.346 ms estimated-time field is the minimum. Hidden/K/V relative L2 versus original FP32 ONNX CPU was 1.351%/0.377%/0.650%, without a next-token quality tolerance. There is no complete 28-layer prefill/decode/KV-ring/sampling loop, resident Omni controller or complete-request timing. | Qualify a complete X Elite resident artifact/controller with state, token quality, memory admission/cancel and sustained power; [X Elite NPU raw evidence](../../../e2e_expansion_20260923/evidence/spark_xelite_full_attention/README.md). | -| PC Snapdragon X Elite CRD / AI Hub | Qwen3-TTS 0.6B CustomVoice | COMPONENT_ONNX_CPU_VOCODER_PARITY_PROFILED; DIRECTML_RUNTIME_FAILED; TALKER_FP16_NUMERIC_FAIL; QNN_NPU_VOCODER_NUMERIC_MEASURED_PROFILE_TIMEOUT; SHORT_QNN_VOCODER_NUMERIC_MEASURED_PLACEMENT_PENDING; TALKER_ONNX_CPU_NUMERIC_PROFILED; NOT_E2E_PROFILEABLE | The fresh ONNX vocoder inference on the exact Windows 11 ARM64 X Elite CRD requested CPU and returned an unsaturated waveform on the pinned fixture, matching local ONNX Runtime CPU with 8.09e-7 relative L2 and 121.84 dB SNR. Its 94-sample component profile had nearest-rank p50/p95 6.384/6.644 s and all 634 execution-detail rows on CPU; 5.917 s estimated time is the sample minimum. This is one component, without a listening-quality gate. Corrected DirectML inference and profile both failed at provider initialization with Windows error 80070057. A separate FP16 QNN DLC vocoder inference returned an unsaturated waveform at 1.138% relative L2 / 38.88 dB SNR versus local ONNX CPU on the same fixture; the NPU node-placement profile timed out during graph preparation with no timing or per-node samples, and no listening-quality tolerance is set. A new [two-frame QNN vocoder candidate](../../../e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/xelite/qnn_short_chunk/README.md) compiled for X Elite and returned an unsaturated waveform at 0.930% relative L2 / 40.63 dB SNR versus pinned local ONNX CPU on one synthetic fixture. Its NPU-requested placement profile remains in progress; inference request alone does not verify NPU nodes or latency. The 28-layer FP16 QNN talker step executed, yet changed CPU top-1 token 80 to 1720 with 8.57% logits relative L2 on a synthetic cache fixture; a separate ONNX CPU-requested talker step preserved CPU token 80 with logits relative L2 2.07e-5 and maximum K/V relative L2 2.18e-5. Its 100-sample component profile had p50/p95 191.414/309.428 ms and all 1,744 execution-detail rows on CPU. The older TFLite vocoder target was rejected on Windows ARM; no complete stream exists. | Complete the short QNN vocoder placement profile and audit CPU talker placement; investigate DirectML failure, then qualify complete local streaming, memory and sustained behavior; [vocoder](../../../e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/README.md) and [talker](../../../e2e_expansion_20260924/evidence/qwen_tts_talker_qualcomm/README.md) evidence. | +| PC Snapdragon X Elite CRD / AI Hub | Qwen3-TTS 0.6B CustomVoice | COMPONENT_ONNX_CPU_VOCODER_PARITY_PROFILED; DIRECTML_RUNTIME_FAILED; TALKER_FP16_NUMERIC_FAIL; QNN_NPU_VOCODER_NUMERIC_MEASURED_PROFILE_TIMEOUT; SHORT_QNN_NPU_VOCODER_NUMERIC_PROFILED_SLOW; TALKER_ONNX_CPU_NUMERIC_PROFILED; NOT_E2E_PROFILEABLE | The fresh ONNX vocoder inference on the exact Windows 11 ARM64 X Elite CRD requested CPU and returned an unsaturated waveform on the pinned fixture, matching local ONNX Runtime CPU with 8.09e-7 relative L2 and 121.84 dB SNR. Its 94-sample component profile had nearest-rank p50/p95 6.384/6.644 s and all 634 execution-detail rows on CPU; 5.917 s estimated time is the sample minimum. This is one component, without a listening-quality gate. Corrected DirectML inference and profile both failed at provider initialization with Windows error 80070057. A separate FP16 QNN DLC vocoder inference returned an unsaturated waveform at 1.138% relative L2 / 38.88 dB SNR versus local ONNX CPU on the same fixture; the NPU node-placement profile timed out during graph preparation with no timing or per-node samples, and no listening-quality tolerance is set. A new [two-frame QNN vocoder candidate](../../../e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/xelite/qnn_short_chunk/README.md) compiled for X Elite and returned an unsaturated waveform at 0.930% relative L2 / 40.63 dB SNR versus pinned local ONNX CPU on one synthetic fixture. Its 100-sample placement profile measured nearest-rank p50/p95 3.662/3.676 s for 0.16 s audio, all 728 execution rows on NPU, 15,855,616 bytes reported peak and median component RTF 22.89. The profile used Workbench-generated same-shape input; the waveform used the retained fixture. This is too slow for playable streaming on the tested artifact. The 28-layer FP16 QNN talker step executed, yet changed CPU top-1 token 80 to 1720 with 8.57% logits relative L2 on a synthetic cache fixture; a separate ONNX CPU-requested talker step preserved CPU token 80 with logits relative L2 2.07e-5 and maximum K/V relative L2 2.18e-5. Its 100-sample component profile had p50/p95 191.414/309.428 ms and all 1,744 execution-detail rows on CPU. The older TFLite vocoder target was rejected on Windows ARM; no complete stream exists. | Find a quality-passing, much faster vocoder path and audit CPU talker placement; investigate DirectML failure, then qualify complete local streaming, memory and sustained behavior; [vocoder](../../../e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/README.md) and [talker](../../../e2e_expansion_20260924/evidence/qwen_tts_talker_qualcomm/README.md) evidence. | | PC Snapdragon X Elite CRD / AI Hub | Qwen3.8-27B | NOT_E2E_PROFILEABLE | No complete qualified target artifact and device-local stateful Omni backend. Hub jobs provide component execution, not an on-device generation/streaming controller; ADB has no attached device. | Supply and validate the missing matching artifact/backend, then run the complete specified workload. | | PC Snapdragon X Elite CRD / AI Hub | MiniCPM-o 4.5 | COMPONENT_SPEECH_HEAD_NPU_NUMERIC_PROFILED; NOT_E2E_PROFILEABLE | The real 20-layer MiniCPM-o speech-head FP16 QNN DLC ran on the exact X Elite CRD. Same-fixture inference returned 41 finite tensors, preserved ONNX CPU top-1 speech token 1867, and had 0.3170% logits relative L2 and 1.221% maximum K/V relative L2. Its 100-sample component profile had nearest-rank p50/p95 9.905/11.037 ms, with all 944 execution-detail rows on NPU. Reported inference peak memory was 79,208,448 bytes and first-load peak 1,081,573,376 bytes. Thinker/encoders/vocoder and a device-local stateful Omni backend remain unqualified. | Validate the complete multimodal pipeline, state, quality, co-resident memory and sustained behavior; [X Elite raw evidence](../../../e2e_expansion_20260924/evidence/minicpmo_tts_qualcomm/README.md). | | PC Snapdragon X Elite CRD / AI Hub | InternVLA-A1 | COMPONENT_COSMOS_NPU_NUMERIC_PROFILED; NOT_E2E_PROFILEABLE | The real Place_Markpen Cosmos FP16 QNN image encoder returned a finite FP32 latent for the six-frame synthetic pattern on exact X Elite CRD. Against the pinned source, relative L2 was 0.0608%, SNR 64.32 dB and cosine 0.999999815. The 100-sample six-frame encoder profile had p50/p95 82.939/83.837 ms, with all 121 execution-detail rows on NPU. Reported component inference peak memory was 24,059,904 bytes and first-load peak 692,891,648 bytes. No real-observation action tolerance, complete policy or local stateful Omni backend exists. | Validate real observations, action units/timing, state, admission, cancellation and sustained behavior; [X Elite Cosmos raw evidence](../../../e2e_expansion_20260924/evidence/internvla_cosmos_qualcomm/README.md). | | Mobile Galaxy S25 / Snapdragon 8 Elite for Galaxy | Spark-X2.5 | COMPONENT_NPU_INFERENCE_PROFILED; NOT_E2E_PROFILEABLE | The real 1.7B checkpoint one-layer W8A16 full-attention decoder artifact executed for a fixed 1,024-token cache bucket. Same-fixture outputs matched S24; the hosted S25 profile had 100 component samples at nearest-rank p50/p95 0.957/1.028 ms with all 73 execution-detail rows on NPU. FP32-source hidden-state relative L2 was 1.351% without a downstream token-quality gate. No full 28-layer prefill/decode, KV/ring, sampling or resident controller ran on S25. | Implement and validate local 128-token generation, 512/1024 transitions, cancellation, resident memory and sustained power; [S25/S24 component evidence](../../../e2e_expansion_20260923/evidence/spark_s24_full_attention/README.md). | -| Mobile Galaxy S25 / Snapdragon 8 Elite for Galaxy | Qwen3-TTS 0.6B CustomVoice | TALKER_FP16_NUMERIC_FAIL; CPU_TALKER_AND_VOCODER_NUMERIC_PROFILED; QNN_VOCODER_MEMORY_FAIL; FULL_TFLITE_MIXED_COMPONENT_PROFILED; PINNED_FULL_TFLITE_GPU_NUMERIC_MEASURED_PLACEMENT_PENDING; SHORT_TFLITE_MIXED_COMPONENT_PROFILED; NOT_E2E_PROFILEABLE | Fresh FP16 predictor inference passed; its one output exactly matches the same compiled artifact/dataset reference. The same historical S25 TFLite vocoder target requested on CPU returned a finite unsaturated waveform at 4.04e-6 relative L2 / 107.86 dB SNR versus original ONNX CPU on the retained fixture. Its 100-sample component profile measured nearest-rank p50/p95 4.418/4.641 s, with all 738 execution-detail rows on CPU and reported inference peak memory 867,893,248 bytes. A separate 28-layer TFLite CPU-requested talker step returned 57 finite outputs, preserved ONNX CPU top-1 speech token 80, and had logits relative L2 3.33e-5 and maximum new-cache relative L2 3.41e-5. Its 100-sample component profile measured p50/p95 30.161/31.301 ms, with all 1,800 execution-detail rows on CPU and reported inference peak memory 1,929,318,400 bytes. The FP16 QNN talker step executed on the synthetic cache fixture but changed CPU top-1 speech token 80 to 412 with 8.99% logits relative L2 and maximum new-K/V relative L2 11.71%. A separately compiled FP16 QNN DLC talker also changed top-1 to 1720 with 8.77% logits relative L2 and 11.27% maximum new-cache relative L2. Both tested NPU talker artifacts are numerically unqualified. A fresh 25-frame FP16 QNN DLC vocoder compiled for S25, but its NPU-requested inference failed when Workbench reported device memory limits; it returned no waveform or profile samples. A [two-frame short-window export](../../../e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/qnn_short_chunk/README.md) matched local ONNX CPU at 2.96e-6 relative L2, then its separately compiled S25 FP16 QNN DLC also failed NPU-requested inference with the same reported device-memory limit. Its local HX370 WSL CPU component p50/p95 was 1.365/1.453 s to produce 0.16 s audio, not an S25 estimate. A separate [two-frame TFLite GPU-requested inference](../../../e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/tflite_short_chunk/README.md) returned finite unsaturated audio at 2.325% relative L2 / 32.67 dB SNR versus local ONNX CPU on one fixture; its 100-sample placement profile measured p50/p95 0.559/1.052 s for 0.16 s audio, with 403 CPU and 335 GPU execution-detail rows and 379,121,664 bytes reported peak. The mixed CPU/GPU short component median RTF is 3.49. A separate [25-frame TFLite GPU-requested profile](../../../e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/full_tflite_gpu_profile/README.md) of the historical target measured 100 component samples at p50/p95 0.839/1.157 s for 2 s audio, with 403 CPU and 335 GPU rows, reported 374,591,488-byte peak and median component RTF 0.420. Its retained GPU waveform differed from ONNX CPU by 4.24% relative L2 on one fixture. A pinned CustomVoice eager 25-frame decoder matched the historical ONNX CPU waveform at 1.89e-6 relative L2 on that fixture, but the historical export revision and listening quality remain unverified. A fresh [revision-pinned 25-frame TFLite export](../../../e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/full_pinned_tflite/README.md) matched local eager CPU at 1.99e-6 relative L2 and compiled for S25. Its GPU-requested inference returned an unsaturated waveform at 4.239% relative L2 / 27.46 dB SNR versus pinned local ONNX CPU, bitwise identical to the historical GPU output on this one fixture. Its own placement profile remains pending; the historical target's timing does not transfer automatically. No complete co-resident NPU/GPU stream exists. | Improve and revalidate the [talker artifact](../../../e2e_expansion_20260924/evidence/qwen_tts_talker_qualcomm/README.md), then qualify the [device-local predictor/vocoder pipeline](../../../e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/README.md), output quality, recovery and 30-minute thermal behavior. | +| Mobile Galaxy S25 / Snapdragon 8 Elite for Galaxy | Qwen3-TTS 0.6B CustomVoice | TALKER_FP16_NUMERIC_FAIL; CPU_TALKER_AND_VOCODER_NUMERIC_PROFILED; QNN_VOCODER_MEMORY_FAIL; FULL_TFLITE_MIXED_COMPONENT_PROFILED; PINNED_FULL_TFLITE_MIXED_COMPONENT_PROFILED; SHORT_TFLITE_MIXED_COMPONENT_PROFILED; NOT_E2E_PROFILEABLE | Fresh FP16 predictor inference passed; its one output exactly matches the same compiled artifact/dataset reference. The same historical S25 TFLite vocoder target requested on CPU returned a finite unsaturated waveform at 4.04e-6 relative L2 / 107.86 dB SNR versus original ONNX CPU on the retained fixture. Its 100-sample component profile measured nearest-rank p50/p95 4.418/4.641 s, with all 738 execution-detail rows on CPU and reported inference peak memory 867,893,248 bytes. A separate 28-layer TFLite CPU-requested talker step returned 57 finite outputs, preserved ONNX CPU top-1 speech token 80, and had logits relative L2 3.33e-5 and maximum new-cache relative L2 3.41e-5. Its 100-sample component profile measured p50/p95 30.161/31.301 ms, with all 1,800 execution-detail rows on CPU and reported inference peak memory 1,929,318,400 bytes. The FP16 QNN talker step executed on the synthetic cache fixture but changed CPU top-1 speech token 80 to 412 with 8.99% logits relative L2 and maximum new-K/V relative L2 11.71%. A separately compiled FP16 QNN DLC talker also changed top-1 to 1720 with 8.77% logits relative L2 and 11.27% maximum new-cache relative L2. Both tested NPU talker artifacts are numerically unqualified. A fresh 25-frame FP16 QNN DLC vocoder compiled for S25, but its NPU-requested inference failed when Workbench reported device memory limits; it returned no waveform or profile samples. A [two-frame short-window export](../../../e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/qnn_short_chunk/README.md) matched local ONNX CPU at 2.96e-6 relative L2, then its separately compiled S25 FP16 QNN DLC also failed NPU-requested inference with the same reported device-memory limit. Its local HX370 WSL CPU component p50/p95 was 1.365/1.453 s to produce 0.16 s audio, not an S25 estimate. A separate [two-frame TFLite GPU-requested inference](../../../e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/tflite_short_chunk/README.md) returned finite unsaturated audio at 2.325% relative L2 / 32.67 dB SNR versus local ONNX CPU on one fixture; its 100-sample placement profile measured p50/p95 0.559/1.052 s for 0.16 s audio, with 403 CPU and 335 GPU execution-detail rows and 379,121,664 bytes reported peak. The mixed CPU/GPU short component median RTF is 3.49. A separate [25-frame TFLite GPU-requested profile](../../../e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/full_tflite_gpu_profile/README.md) of the historical target measured 100 component samples at p50/p95 0.839/1.157 s for 2 s audio, with 403 CPU and 335 GPU rows, reported 374,591,488-byte peak and median component RTF 0.420. Its retained GPU waveform differed from ONNX CPU by 4.24% relative L2 on one fixture. A pinned CustomVoice eager 25-frame decoder matched the historical ONNX CPU waveform at 1.89e-6 relative L2 on that fixture, but the historical export revision and listening quality remain unverified. A fresh [revision-pinned 25-frame TFLite export](../../../e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/full_pinned_tflite/README.md) matched local eager CPU at 1.99e-6 relative L2 and compiled for S25. Its GPU-requested inference returned an unsaturated waveform at 4.239% relative L2 / 27.46 dB SNR versus pinned local ONNX CPU, bitwise identical to the historical GPU output on this one fixture. Its own 100-sample placement profile measured nearest-rank p50/p95 0.671/1.454 s for 2 s audio, median component RTF 0.335, with 403 CPU and 335 GPU execution rows and 380,100,608 bytes reported peak memory. The profile used Workbench-generated same-shape input; the numerical inference used the retained fixture. The historical 0.839/1.157 s profile is a separate run. A [real generated-code 25-frame window](../../../e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/README.md) passed local ONNX/eager parity at 1.02e-6 relative L2 and matched the corresponding full FP32 decoder segment at 7.26e-8 relative L2; the same pinned S25 GPU-requested target returned an unsaturated waveform at 1.153% relative L2 / 38.76 dB SNR versus local ONNX CPU on this generated-code window, without a listening-quality gate. No complete co-resident NPU/GPU stream exists. | Improve and revalidate the [talker artifact](../../../e2e_expansion_20260924/evidence/qwen_tts_talker_qualcomm/README.md), then qualify the [device-local predictor/vocoder pipeline](../../../e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/README.md), output quality, recovery and 30-minute thermal behavior. | | Mobile Galaxy S25 / Snapdragon 8 Elite for Galaxy | Qwen3.8-27B | NOT_E2E_PROFILEABLE | No complete qualified target artifact and device-local stateful Omni backend. Hub jobs provide component execution, not an on-device generation/streaming controller; ADB has no attached device. | Supply and validate the missing matching artifact/backend, then run the complete specified workload. | | Mobile Galaxy S25 / Snapdragon 8 Elite for Galaxy | MiniCPM-o 4.5 | COMPONENT_FP16_SPEECH_HEAD_NPU_NUMERIC_PROFILED; NOT_E2E_PROFILEABLE | Fresh FP16 speech-head inference passed; all 41 outputs exactly matched the same artifact/dataset reference. Against the recovered original ONNX CPU source on the same fixture, logits relative L2 was 0.3605%, maximum new-K/V relative L2 1.311%, and top-1 speech token 1867 matched. Historical FP16 parity passed; tested W8A16 top-1 parity failed. The 100-sample component profile had p50/p95 7.197/7.615 ms, with all 1,029 execution-detail rows on NPU and reported inference peak memory 185,991,168 bytes. Generated-audio quality and qualified thinker/encoders/vocoder remain unverified. | Qualify a complete mobile checkpoint/export and the full multimodal/stateful pipeline; [speech-head evidence](../../../e2e_expansion_20260924/evidence/minicpmo_tts_qualcomm/README.md). | | Mobile Galaxy S25 / Snapdragon 8 Elite for Galaxy | InternVLA-A1 | COMPONENT_COSMOS_NPU_NUMERIC_PROFILED; SYNTHETIC_ACTION_SENSITIVITY_MEASURED; NOT_E2E_PROFILEABLE | The real Place_Markpen Cosmos image encoder FP32 ONNX export matched its source on two six-frame fixtures. Galaxy S25 FP16 QNN DLC inference returned a finite latent within 0.0524% relative L2 / 65.61 dB SNR of the pinned pattern source. Injecting this device latent into the same real policy on local CPU changed synthetic `[1,50,32]` actions by 0.742% relative L2; the CPU source repeated bitwise identically. The 100-sample six-frame encoder profile had nearest-rank p50/p95 49.665/51.382 ms, all 121 execution-detail rows on NPU, and a reported inference peak of 158,232,576 bytes. No real-observation task tolerance exists, and the action policy was not resident on S25. ADB has no attached device. | Extend the [S25 encoder component](../../../e2e_expansion_20260924/evidence/internvla_cosmos_qualcomm/README.md), then validate real action quality, device-local state, admission, cancellation and sustained behavior. | diff --git a/docs/design/omni_edge_e2e_roadmap.md b/docs/design/omni_edge_e2e_roadmap.md index 3a5d8e8b2..4bba3fa66 100644 --- a/docs/design/omni_edge_e2e_roadmap.md +++ b/docs/design/omni_edge_e2e_roadmap.md @@ -48,8 +48,16 @@ now matches local eager CPU at 1.99e-6 relative L2 and has compiled for S25 TFLite. Its GPU-requested inference returned an unsaturated waveform at 4.239% relative L2 versus pinned local ONNX CPU, bitwise identical to the historical target's output on the same fixture. The new target's own -node-placement profile is pending, so the historical timing does not yet -qualify this revision-pinned artifact. +100-sample node-placement profile measured p50/p95 0.671/1.454 s for 2 s +audio, with 403 CPU and 335 GPU rows (median component RTF 0.335). This +qualifies only the tested vocoder component; full co-resident streaming, +speech quality and sustained behavior remain open. +A [real generated-code 25-frame window](../../benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s25/real_code_window/README.md) +also passed local ONNX/eager parity at 1.02e-6 relative L2 and matched its +full FP32 decoder segment at 7.26e-8 relative L2. The corresponding S25 +GPU-requested inference returned an unsaturated waveform at 1.153% relative +L2 / 38.76 dB SNR versus ONNX CPU. This tests one generated-code window, +not mobile talker/vocoder co-residency or speech quality. On [Galaxy S24](../../benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/s24/tflite_short_chunk/README.md), the same pinned two-frame TFLite source passed CPU-requested waveform parity at 5.75e-6 relative L2, but the GPU-requested waveform saturated completely @@ -59,8 +67,10 @@ measured 0.601/0.895 s but failed the waveform gate. A FP32-preserving GPU option returned the same saturated waveform bitwise, so it did not repair the tested delegate artifact. A [Snapdragon X Elite short QNN graph](../../benchmarks/edge_harness/results/e2e_expansion_20260924/evidence/qwen_tts_vocoder_qualcomm/xelite/qnn_short_chunk/README.md) compiled and returned an unsaturated NPU-requested waveform at 0.930% -relative L2 versus local ONNX CPU on one synthetic fixture. Its node-placement -profile is pending. Neither changes M2 support +relative L2 versus local ONNX CPU on one synthetic fixture. Its 100-sample +profile attributed all 728 rows to NPU, but p50/p95 was 3.662/3.676 s for +0.16 s audio (median component RTF 22.89). This artifact therefore fails +the playable-stream latency gate. Neither changes M2 support depth before full stream and quality gates. The [Spark hosted Qualcomm component expansion](../../benchmarks/edge_harness/results/e2e_expansion_20260923/evidence/spark_s24_full_attention/README.md) advances M1 only to one real fixed-shape decoder attention layer. The same W8A16 QNN DLC ran on S24, Snapdragon X Elite CRD and SA8775P ADP NPUs with identical outputs on one pinned fixture; S25 had prior separate component evidence. On [RB3 Gen 2](../../benchmarks/edge_harness/results/e2e_expansion_20260923/evidence/spark_rb3_full_attention/README.md), the tested W8A16 graph failed QNN loading even after an exact-device compile. FP32 ONNX ran on CPU with near-source parity, while calibrated W8A8 ran on NPU but had 20.9% hidden-state relative L2 and remains numerically unqualified. These are component C results, not full M1: device-local 28-layer prefill/continuous decode, KV/ring, sampling, token quality, admission/cancellation, complete-request timing and sustained power/thermal gates remain open. AI Hub is the test facility, not a deployment dependency.