tzhouam and Claude Fable 5.1
285db2fa3c
edge results: WP1 edge-profile stream latency (0.6B/1.7B) + init timelines
...
1.7B: TTFA p50 43.9 ms, playback_start 43.9 ms, stall 0 (baseline 32.0 / 116.2 / 84.7).
0.6B: TTFA p50 46.1 ms, playback_start 46.1 ms, stall 0 (baseline 27.8 / 95.4 / 67.3).
RTF unchanged (~0.1). Measured on L20X via gpu run --gpus 1 (canhazgpu).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com >
2026-09-09 11:21:48 +08:00
tzhouam and Claude Fable 5.1
35e43b9983
[Edge][WP3] Add CPU OmniPlatform (AR + generation stages on vLLM's CPU backend)
...
* vllm_omni/platforms/cpu/platform.py: CPUOmniPlatform(OmniPlatform, CpuPlatform)
selected when VLLM_TARGET_DEVICE=cpu or a +cpu vLLM wheel is installed
(builtin plugin "cpu"; explicit CPU target short-circuits accelerator probes
like vLLM does). NUMA-based memory queries, one logical device, no graph
wrapper (sentinel type for isinstance checks), inductor opt-in via
VLLM_OMNI_CPU_INDUCTOR=1, diffusion stages unsupported.
* vllm_omni/platforms/cpu/worker: CPUARWorker / CPUGenerationWorker compose
vLLM's CPUWorker with the Omni mixin (XPU pattern); CPUARModelRunner /
CPUGenerationModelRunner compose the Omni runners with CPUModelRunner
(no graphs, no async omni output; sleep/wake return error ACKs).
* patch.py skips the CuMem patch on CPU-only torch; Qwen3-TTS speaker encoder
and prompt-embeds dtype follow the model dtype (float32/float16 on CPU).
* vllm_omni/deploy/qwen3_tts.yaml: `platforms: cpu:` overlay (eager, no async
scheduling, absolute kv_cache_memory_bytes, disjoint OMP thread ranges).
* requirements/cpu.txt (+psutil, intel-openmp), docs/getting_started/installation/cpu.md.
Measured (this x86 Xeon 8480C host, vllm 0.28.0+cpu, bf16, 12+4 threads):
Qwen3-TTS 0.6B CustomVoice end-to-end on CPU, 12 prompts: TTFA p50 292 ms,
RTF ~1.7 (25-frame chunk ~4.6 s), init 185 s. Result JSON + init timeline
committed under benchmarks/tts/edge_results/.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com >
2026-09-09 11:02:14 +08:00
tzhouam and Claude Fable 5.1
f804bbd679
[Edge][WP0] Record Qwen3-TTS streaming baselines on 1x L20X
...
Default deploy config (1 -> 25 frame schedule), 12 prompts x 3 repeats,
warm-up excluded, one GPU via `gpu run`:
1.7B CustomVoice: TTFA p50 32.0 ms, playback-start p50 116 ms (stall at
TTFA 85 ms), RTF_total 0.086, chunk gap p50 163 ms, init 271 s (cold caches)
0.6B CustomVoice: TTFA p50 27.8 ms, playback-start p50 95 ms (stall 67 ms),
RTF_total 0.077, chunk gap p50 146 ms, init 152 s
Per-request chunk arrivals, memory and the full summaries are in the JSON
(force-added: *.json is gitignored repo-wide).
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com >
2026-09-09 10:49:37 +08:00
tzhouam and Claude Fable 5.1
b70207f57f
[Edge][WP0] Add offline streaming-latency benchmark for chunked TTS
...
Adds benchmarks/tts/stream_latency_bench.py (AsyncOmni driver reusing the
Qwen3-TTS example's prompt builder; --model override, warm-up/repeat,
--init-only, per-chunk arrivals, peak RSS and per-process GPU memory) and
benchmarks/tts/stream_latency_metrics.py (TTFA, underrun-free playback start,
stall at TTFA, RTF streamed/total, chunk gaps, percentiles) with CPU unit
tests. Results go under benchmarks/tts/edge_results/stream_latency/.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com >
2026-09-09 10:24:53 +08:00