Profile Qwen3-TTS hybrid Radeon and CPU route

This commit is contained in:
tzhouam committed 2026-09-23 19:59:04 +08:00
1 parent 54c086acae
commit 4fee1dbe39
54 files changed
+5323 -6

No files matched your search

+1
View File
@@ -108,6 +108,7 @@ Model support is specific to the following pairings:
| Spark-X2.5-1.7B INT8 | CPU execution; no NVIDIA dependency | Native Windows | Model path not yet qualified; CPU Add graph success is not text-model support. |
| Qwen3-TTS-12Hz-0.6B CustomVoice | CPU + RTX 5090 Laptop GPU; both stages on GPU | WSL / vLLM CUDA | Current branch: real-weight audio streaming completes; playback stalls and remaining M2 gates stay open. |
| Qwen3-TTS-12Hz-0.6B CustomVoice | CPU + RTX 5090 Laptop GPU; both stages on GPU | Native Windows / vLLM CUDA | Current branch: real-weight audio streaming completes; playback stalls and remaining M2 gates stay open. |
| Qwen3-TTS-12Hz-0.6B CustomVoice, Q8_0 GGUF talker + F16 codec | CPU + Radeon 890M iGPU; code predictor on CPU FP32, talker/codec on Radeon | Native Windows / standalone CrispASR Vulkan | Two named text-to-WAV requests and a 20-request serial resident profile passed; [placement and quality evidence](../../benchmarks/edge_harness/results/e2e_expansion_20260923/evidence/qwen_tts_radeon890m_joint/README.md). Still outside Omni, slower than real time and unqualified for streaming or sustained use. |
| Qwen3.8-27B vision tower, FP32 ONNX | CPU + Radeon 890M iGPU | Windows / ORT DirectML | Historical component validation only; not the complete 27B model or a newly qualified graph-stage vision-to-language pipeline. |
| Qwen3.8-27B vision tower, tested A16W8 exports | CPU + AMD NPU | Windows / ORT VitisAI | Tested whole-tower artifacts were rejected by NPU partitioning. This does not rule out other exports or smaller components. |
| Qwen3.8-27B complete multimodal pipeline | CPU/iGPU/NPU, with optional discrete GPU | Target-specific | Fresh public Omni startup fails pipeline resolution on Windows/WSL; earlier bare-vLLM short-text loading is separate and not multimodal validation. |
+11
View File
@@ -56,6 +56,17 @@ documents the isolated Transformers 4.57.3 dependency and short process-memory
trace. Omni CPU stage binding, playable streaming, loading-peak admission and
sustained power/thermal tests remain M2 work.
A separate native Windows CPU+Radeon 890M Qwen3-TTS route now completes two
named text-to-WAV requests in standalone CrispASR using a Q8_0 talker, F16
codec, Vulkan talker/codec and CPU FP32 code predictor. Whisper tiny.en returned
both reference sentences exactly; one warmup and 20 serial resident-server
requests measured p50/p95 3.310/3.345 s for 2.56 s of audio. The
[placement and failed-route record](../../benchmarks/edge_harness/results/e2e_expansion_20260923/evidence/qwen_tts_radeon890m_joint/README.md)
shows that default Vulkan silently becomes CPU for this backend, all-Vulkan
generation ran away, and DirectML did not pass quality. The explicit hybrid
route remains outside Omni and slower than real time. Its CPU+iGPU evidence
does not advance the NPU joint-execution row or mobile M2 gate.
## Architecture and prerequisites
Keep Omni's PipelineConfig, StageRuntime, StageClient, orchestration and admission