Skip to content

Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.

Archive merged Transformers Qwen3-ASR fix (#3342)

zhifu gao69071c8bab4b
diff --git a/docs/community_growth_20k.md b/docs/community_growth_20k.md
index af4ec7da3..baac677ad 100644
--- a/docs/community_growth_20k.md
+++ b/docs/community_growth_20k.md
@@ -34,13 +34,13 @@ This plan focuses on useful adoption work rather than vanity marketing: if more
## Current campaign snapshot
-As of 2026-07-22 19:20 UTC, the ecosystem has 35,786 combined GitHub stars, or 4,562 additional stars since the 31,224 baseline. The remaining gap to the +20,000 target is 15,438 stars by 2026-09-30, which requires roughly 221 stars/day across the remaining 70 days.
+As of 2026-07-22 19:33 UTC, the ecosystem has 35,786 combined GitHub stars, or 4,562 additional stars since the 31,224 baseline. The remaining gap to the +20,000 target is 15,438 stars by 2026-09-30, which requires roughly 221 stars/day across the remaining 70 days.
| Repository | Stars | Forks | Open issues | Open PRs | Last push |
|---|---:|---:|---:|---:|---|
| `modelscope/FunASR` | 19,407 | 1,952 | 2 | 0 | 2026-07-22 |
| `FunAudioLLM/Fun-ASR` | 1,416 | 139 | 0 | 0 | 2026-07-22 |
-| `FunAudioLLM/SenseVoice` | 8,916 | 795 | 0 | 0 | 2026-07-22 |
+| `FunAudioLLM/SenseVoice` | 8,916 | 796 | 0 | 0 | 2026-07-22 |
| `modelscope/FunClip` | 6,047 | 721 | 0 | 0 | 2026-07-22 |
Keep this snapshot fresh during weekly planning. The ecosystem mode also reports the remaining gap, days left to 2026-09-30, and the required daily average:
@@ -212,7 +212,7 @@ High-star feature requests and roadmap issues are earlier in the funnel than PRs
|---|---|---|
| `Kedreamix/Linly-Talker#151` FunASR link refresh | Keeps a 3.3k-star digital-avatar conversational system's ASR docs pointed at the current official FunASR repository, so users comparing Whisper and FunASR land on maintained setup docs | PR updates the root README and `ASR/README.md` references from `github.com/alibaba-damo-academy/FunASR` to `github.com/modelscope/FunASR`; monitor for maintainer acceptance without extra pings. |
| `huggingface/transformers#46180` Fun-ASR-Nano model support | Makes Fun-ASR-Nano usable through the default HF API surface and model docs | Head `5efce76908` is ready for maintainer review: GitHub reports the PR mergeable, all active checks are successful or intentionally skipped, and the only current unresolved review thread is a non-code architecture discussion where the maintainer is leaving `FunAsrNanoModel(AudioFlamingo3Model)` open for other reviewer thoughts because the adaptor is larger than a typical projector. The final review cleanup keeps feature extraction in Torch through `self.pad`, uses inherited batch-capable `decode`, removes the encoder-specific initializer, and preserves the pinned H100 checkpoint evidence for single and batch Chinese/English transcription with zero missing, unexpected, or mismatched keys. Avoid duplicate pings unless a new actionable review thread or failing check appears. |
-| `huggingface/transformers#47111` Qwen3-ASR hotword and language parsing fixes | Keeps the high-visibility Transformers Qwen3-ASR implementation aligned with upstream hotword/context behavior, language hints, and processor training paths used by downstream Fun-ASR-Nano/Qwen3-ASR adopters | The Qwen3-ASR slow job was explicitly rerun and reported no PR-specific failures on head `caf37300fc3d3333c3d7c2162a92a886e424eb1c`; the aggregate CI recap still has unrelated suite failures. Monitor for reviewer requests around hotword serialization, language parsing, and qwen3_asr processor tests before adding any FunASR-side comment. |
+| `huggingface/transformers#47111` Qwen3-ASR hotword and language parsing fixes | Keeps the high-visibility Transformers Qwen3-ASR implementation aligned with upstream hotword/context behavior, language hints, and processor training paths used by downstream Fun-ASR-Nano/Qwen3-ASR adopters | Merged on 2026-07-22 as `c7f9c8815610d27e41a6b0b0cc9e2d3c49468d1d`. The merged change adds Qwen3-ASR custom hotword support, language parsing fixes, processor training-path tests, conversion updates, and documentation examples, so treat this as a completed Transformers-side ASR hardening win and watch the next Transformers release/model docs rather than keeping it in the active review queue. |
| `sgl-project/sglang-omni#1078` Fun-ASR serving support | Exposes Fun-ASR-Nano through a high-visibility serving runtime for ASR benchmarks and GPU deployment | Merged on 2026-07-18 after replacing the closed #898 branch. The merged PR includes Fun-ASR model runtime support, OpenAI-compatible serving, cookbook docs, SeedTTS benchmark tasks, unit tests, and review fixes from #1079. Public H200 SeedTTS EN full-set validation reports 1,088/1,088 samples, 0 failed, 1.71% corpus WER, 117.37 samples/s at concurrency 32, mean latency 0.271s, and mean RTF 0.0587; keep watching for follow-up issues rather than pinging the closed PR. |
| `vllm-project/vllm#42478` Qwen3-ASR streaming postprocessing | Improves the upstream Qwen3-ASR streaming path used by OpenAI-compatible transcription clients, so downstream apps can consume clean SSE transcript deltas without model-specific `language ...<asr_text>` cleanup | Merged with maintainer approval. Treat this as a completed downstream ASR serving hardening win; watch vLLM release behavior for CJK no-space streaming regressions, but do not keep the closed PR in the active operator queue. |
| `vllm-project/vllm#47729` MOSS-Transcribe-Diarize support | Expands vLLM's OpenAI-compatible `/v1/audio/transcriptions` route for long-form ASR with timestamps and speaker labels, creating another high-visibility comparison point for Qwen3-ASR / Fun-ASR-Nano-style serving behavior | Merged with all checks green. The non-blocking downstream ASR contract note remains useful for response-shape monitoring: `response_format=json` should keep the plain `{ "text": "..." }` response stable and any future `segments` should remain additive. |