docs: make native Transformers the Nano starting point

.github/workflows/native-transformers-examples.yml

@@ -0,0 +1,28 @@
+name: Native Transformers examples
+
+on:
+  pull_request:
+    paths: ['README*.md', 'examples/**', 'tests/**', '.github/workflows/native-transformers-examples.yml']
+  push:
+    branches: [main, 'codex/transformers-first-*']
+    paths: ['README*.md', 'examples/**', 'tests/**', '.github/workflows/native-transformers-examples.yml']
+
+permissions:
+  contents: read
+
+jobs:
+  offline-contracts:
+    runs-on: ubuntu-latest
+    timeout-minutes: 10
+    steps:
+      - uses: actions/checkout@v4
+      - uses: actions/setup-python@v5
+        with:
+          python-version: '3.12'
+      - name: Install model-free test dependencies
+        run: python -m pip install 'numpy==1.26.4' 'librosa==0.11.0' 'soundfile==0.13.1' 'pytest==8.4.2'
+      - name: Validate examples and documentation without downloading weights
+        run: >-
+          python -m pytest -q tests/test_transformers_quickstart.py
+          tests/test_examples_smoke.py tests/test_funasr_requirement.py
+          tests/test_timestamp_documentation.py tests/test_moss_ecosystem_docs.py

README.md

@@ -23,7 +23,7 @@ Model repositories: **Fun-ASR-Nano** ([ModelScope](https://www.modelscope.cn/mod
 Online Experience:
 [ModelScope Community Space](https://modelscope.cn/studios/FunAudioLLM/Fun-ASR-Nano), [huggingface space](https://huggingface.co/spaces/FunAudioLLM/Fun-ASR-Nano)
 
-[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/QwenAudio/Fun-ASR/blob/main/examples/colab/fun_asr_nano_quickstart.ipynb)
+[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/QwenAudio/Fun-ASR/blob/main/examples/colab/fun_asr_nano_transformers.ipynb)
 
 [Runnable examples](examples/README.md) cover quickstart inference, direct inference, speaker diarization, vLLM batch inference, and the streaming SDK.
 
@@ -38,10 +38,50 @@ Online Experience:
 
 # What's New 🔥
 
-- **FunASR 1.4.14** is the current Python release for source installs, MOSS discovery, and realtime or industrial deployment. Install with `python -m pip install -U "funasr==1.4.14"`. [Release ->](https://github.com/modelscope/FunASR/releases/tag/v1.4.14)
+- **FunASR 1.4.15** is the current Python release for source installs, MOSS discovery, and realtime or industrial deployment. Install with `python -m pip install -U "funasr==1.4.15"`. [Release ->](https://github.com/modelscope/FunASR/releases/tag/v1.4.15)
 - **MOSS-Transcribe-Diarize** is a third-party OpenMOSS model for offline long-form transcription, timestamps, and anonymous speaker labels, with FunASR service, Docker, Kubernetes, vLLM, SGLang, LocalAI, and FunClip deployment paths. [Deploy MOSS ->](https://www.funasr.com/deploy/moss-transcribe-diarize.html)
 - **Production deployment** covers realtime WebSocket serving, native vLLM batch/streaming paths, and verified llama.cpp / GGUF packages for Linux, macOS, and Windows. [Runtime v0.2.6 ->](https://github.com/modelscope/FunASR/releases/tag/runtime-llamacpp-v0.2.6) · [vLLM guide ->](docs/vllm_guide.md)
 
+# Native Transformers quickstart
+
+Transcribe with the released Transformers 5.17.0 package. No toolkit installation or remote Python code is needed. Base Nano supports Chinese, English and Japanese; the 31-language MLT checkpoint is separate.
+
+```bash
+python -m pip install 'transformers==5.17.0' 'torch==2.10.0' 'torchaudio==2.10.0' 'librosa==0.11.0' 'soundfile==0.13.1'
+```
+
+[Full Python recipe](https://www.funasr.com/en/docs/native-transformers.html) · [Local audio, batches and keywords](examples/transformers/) · [Notebook](examples/colab/fun_asr_nano_transformers.ipynb) · [Space](https://huggingface.co/spaces/FunAudioLLM/Fun-ASR-Nano)
+
+```python
+import torch
+from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor
+
+torch.set_num_threads(4)
+model_id = "FunAudioLLM/Fun-ASR-Nano-2512-hf"
+revision = "d93b302ee7fd505e1b3576120fc142fc6f7820e1"
+audio = "https://huggingface.co/FunAudioLLM/Fun-ASR-Nano-2512/resolve/272c57b82523ada6fd87095e955f8e29100979ab/example/en.mp3"
+
+processor = AutoProcessor.from_pretrained(
+    model_id, revision=revision, trust_remote_code=False, token=False
+)
+model = AutoModelForSpeechSeq2Seq.from_pretrained(
+    model_id, revision=revision, trust_remote_code=False, token=False,
+    dtype=torch.float32,
+).to("cpu").eval()
+inputs = processor.apply_transcription_request(
+    audio=audio, language="en",
+    processor_kwargs={
+        "return_tensors": "pt",
+        "audio_kwargs": {"sampling_rate": 16000},
+        "text_kwargs": {"padding": True},
+    },
+)
+with torch.inference_mode():
+    generated = model.generate(**inputs, max_new_tokens=128, do_sample=False)
+new_tokens = generated[:, inputs.input_ids.shape[1]:]
+print(processor.batch_decode(new_tokens, skip_special_tokens=True)[0])
+```
+
 # Core Features 🎯
 
 **Fun-ASR** focuses on high-precision speech recognition, checkpoint-specific multilingual support, and industry customization capabilities.

README_ja.md

@@ -2,7 +2,7 @@
 
 「[简体中文](README_zh.md)」|「[English](README.md)」|「日本語」
 
-> **FunASR 1.4.14:** 現在の Python リリースで、source install、MOSS の導線、realtime / industrial deployment を提供します。`python -m pip install -U "funasr==1.4.14"`。[Release ->](https://github.com/modelscope/FunASR/releases/tag/v1.4.14) · [Runtime v0.2.6 ->](https://github.com/modelscope/FunASR/releases/tag/runtime-llamacpp-v0.2.6)
+> **FunASR 1.4.15:** 現在の Python リリースで、source install、MOSS の導線、realtime / industrial deployment を提供します。`python -m pip install -U "funasr==1.4.15"`。[Release ->](https://github.com/modelscope/FunASR/releases/tag/v1.4.15) · [Runtime v0.2.6 ->](https://github.com/modelscope/FunASR/releases/tag/runtime-llamacpp-v0.2.6)
 
 > **MOSS-Transcribe-Diarize:** OpenMOSS の第三者モデルで、オフライン長時間転写、timestamp、匿名 speaker label を一度に処理します。FunASR service、Docker、Kubernetes、vLLM、SGLang、LocalAI、FunClip のデプロイパスを利用できます。[MOSS をデプロイ ->](https://www.funasr.com/deploy/moss-transcribe-diarize.html)
 
@@ -27,7 +27,7 @@ Fun-ASRは通義実験室が開発したエンドツーエンド音声認識モ
 オンラインデモ：
 [ModelScope Space](https://modelscope.cn/studios/FunAudioLLM/Fun-ASR-Nano)、[HuggingFace Space](https://huggingface.co/spaces/FunAudioLLM/Fun-ASR-Nano)
 
-[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/QwenAudio/Fun-ASR/blob/main/examples/colab/fun_asr_nano_quickstart.ipynb)
+[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/QwenAudio/Fun-ASR/blob/main/examples/colab/fun_asr_nano_transformers.ipynb)
 
 [実行可能なサンプル](examples/README.md) では、クイックスタート推論、直接推論、話者分離、vLLM バッチ推論、Streaming SDK を確認できます。
 
@@ -40,6 +40,46 @@ Fun-ASRは通義実験室が開発したエンドツーエンド音声認識モ
 
 CPU/エッジ端末では、Fun-ASR-Nano を llama.cpp / GGUF ランタイムで単一バイナリとして実行できます（Python/GPU 不要、内蔵 FSMN-VAD）。[funasr.com/llama-cpp](https://www.funasr.com/llama-cpp.html) · [Nano GGUF](https://huggingface.co/FunAudioLLM/Fun-ASR-Nano-GGUF) · [FSMN-VAD GGUF](https://huggingface.co/FunAudioLLM/fsmn-vad-GGUF)
 
+# Transformers ネイティブクイックスタート
+
+リリース済み Transformers 5.17.0 で音声を文字起こしできます。FunASR toolkit やリモート Python コードは不要です。Nano は中国語・英語・日本語に対応し、31 言語の MLT は別 checkpoint です。
+
+```bash
+python -m pip install 'transformers==5.17.0' 'torch==2.10.0' 'torchaudio==2.10.0' 'librosa==0.11.0' 'soundfile==0.13.1'
+```
+
+[Python ガイド](https://www.funasr.com/en/docs/native-transformers.html) · [ローカル音声・バッチ・キーワード](examples/transformers/) · [Notebook](examples/colab/fun_asr_nano_transformers.ipynb) · [Space](https://huggingface.co/spaces/FunAudioLLM/Fun-ASR-Nano)
+
+```python
+import torch
+from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor
+
+torch.set_num_threads(4)
+model_id = "FunAudioLLM/Fun-ASR-Nano-2512-hf"
+revision = "d93b302ee7fd505e1b3576120fc142fc6f7820e1"
+audio = "https://huggingface.co/FunAudioLLM/Fun-ASR-Nano-2512/resolve/272c57b82523ada6fd87095e955f8e29100979ab/example/en.mp3"
+
+processor = AutoProcessor.from_pretrained(
+    model_id, revision=revision, trust_remote_code=False, token=False
+)
+model = AutoModelForSpeechSeq2Seq.from_pretrained(
+    model_id, revision=revision, trust_remote_code=False, token=False,
+    dtype=torch.float32,
+).to("cpu").eval()
+inputs = processor.apply_transcription_request(
+    audio=audio, language="en",
+    processor_kwargs={
+        "return_tensors": "pt",
+        "audio_kwargs": {"sampling_rate": 16000},
+        "text_kwargs": {"padding": True},
+    },
+)
+with torch.inference_mode():
+    generated = model.generate(**inputs, max_new_tokens=128, do_sample=False)
+new_tokens = generated[:, inputs.input_ids.shape[1]:]
+print(processor.batch_decode(new_tokens, skip_special_tokens=True)[0])
+```
+
 <a name="主要機能"></a>
 
 # 主要機能 🎯

README_ko.md

@@ -2,7 +2,7 @@
 
 「[简体中文](README_zh.md)」|「[English](README.md)」|「[日本語](README_ja.md)」|「한국어」
 
-> **FunASR 1.4.14:** 현재 Python 릴리스로 source install, MOSS 탐색 경로, realtime / industrial deployment를 제공합니다. `python -m pip install -U "funasr==1.4.14"`. [Release ->](https://github.com/modelscope/FunASR/releases/tag/v1.4.14) · [Runtime v0.2.6 ->](https://github.com/modelscope/FunASR/releases/tag/runtime-llamacpp-v0.2.6)
+> **FunASR 1.4.15:** 현재 Python 릴리스로 source install, MOSS 탐색 경로, realtime / industrial deployment를 제공합니다. `python -m pip install -U "funasr==1.4.15"`. [Release ->](https://github.com/modelscope/FunASR/releases/tag/v1.4.15) · [Runtime v0.2.6 ->](https://github.com/modelscope/FunASR/releases/tag/runtime-llamacpp-v0.2.6)
 
 > **MOSS-Transcribe-Diarize:** OpenMOSS의 서드파티 모델로 오프라인 장시간 전사, timestamp, 익명 speaker label을 한 번에 처리합니다. FunASR service, Docker, Kubernetes, vLLM, SGLang, LocalAI, FunClip 배포 경로를 사용할 수 있습니다. [MOSS 배포 ->](https://www.funasr.com/deploy/moss-transcribe-diarize.html)
 
@@ -27,7 +27,7 @@ Fun-ASR는 통의(Tongyi) 실험실에서 개발한 엔드투엔드 음성 인
 온라인 체험:
 [ModelScope Space](https://modelscope.cn/studios/FunAudioLLM/Fun-ASR-Nano), [HuggingFace Space](https://huggingface.co/spaces/FunAudioLLM/Fun-ASR-Nano)
 
-[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/QwenAudio/Fun-ASR/blob/main/examples/colab/fun_asr_nano_quickstart.ipynb)
+[![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/github/QwenAudio/Fun-ASR/blob/main/examples/colab/fun_asr_nano_transformers.ipynb)
 
 [실행 가능한 예제](examples/README.md)는 quickstart 추론, 직접 추론, 화자 분리, vLLM 배치 추론, Streaming SDK를 다룹니다.
 
@@ -40,6 +40,46 @@ Fun-ASR는 통의(Tongyi) 실험실에서 개발한 엔드투엔드 음성 인
 
 CPU/엣지 환경에서는 Fun-ASR-Nano를 llama.cpp / GGUF 런타임으로 단일 바이너리 실행할 수 있습니다(Python/GPU 불필요, FSMN-VAD 내장). 이 GGUF 경로는 Nano의 중국어·영어·일본어 및 중국어 방언 범위에 해당하며, 한국어 인식은 위의 MLT-Nano/FunASR GPU 경로를 사용하세요. [funasr.com/llama-cpp](https://www.funasr.com/llama-cpp.html) · [Nano GGUF](https://huggingface.co/FunAudioLLM/Fun-ASR-Nano-GGUF) · [FSMN-VAD GGUF](https://huggingface.co/FunAudioLLM/fsmn-vad-GGUF)
 
+# Transformers 네이티브 빠른 시작
+
+정식 Transformers 5.17.0으로 음성을 전사합니다. FunASR toolkit 설치나 원격 Python 코드가 필요 없습니다. 기본 Nano는 중국어·영어·일본어를 지원하며 31개 언어의 MLT는 별도 checkpoint입니다.
+
+```bash
+python -m pip install 'transformers==5.17.0' 'torch==2.10.0' 'torchaudio==2.10.0' 'librosa==0.11.0' 'soundfile==0.13.1'
+```
+
+[Python 가이드](https://www.funasr.com/en/docs/native-transformers.html) · [로컬 오디오·배치·키워드](examples/transformers/) · [Notebook](examples/colab/fun_asr_nano_transformers.ipynb) · [Space](https://huggingface.co/spaces/FunAudioLLM/Fun-ASR-Nano)
+
+```python
+import torch
+from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor
+
+torch.set_num_threads(4)
+model_id = "FunAudioLLM/Fun-ASR-Nano-2512-hf"
+revision = "d93b302ee7fd505e1b3576120fc142fc6f7820e1"
+audio = "https://huggingface.co/FunAudioLLM/Fun-ASR-Nano-2512/resolve/272c57b82523ada6fd87095e955f8e29100979ab/example/en.mp3"
+
+processor = AutoProcessor.from_pretrained(
+    model_id, revision=revision, trust_remote_code=False, token=False
+)
+model = AutoModelForSpeechSeq2Seq.from_pretrained(
+    model_id, revision=revision, trust_remote_code=False, token=False,
+    dtype=torch.float32,
+).to("cpu").eval()
+inputs = processor.apply_transcription_request(
+    audio=audio, language="en",
+    processor_kwargs={
+        "return_tensors": "pt",
+        "audio_kwargs": {"sampling_rate": 16000},
+        "text_kwargs": {"padding": True},
+    },
+)
+with torch.inference_mode():
+    generated = model.generate(**inputs, max_new_tokens=128, do_sample=False)
+new_tokens = generated[:, inputs.input_ids.shape[1]:]
+print(processor.batch_decode(new_tokens, skip_special_tokens=True)[0])
+```
+
 <a name="주요-기능"></a>
 
 # 주요 기능 🎯

README_zh.md

@@ -36,10 +36,50 @@ Fun-ASR 是通义实验室推出的端到端语音识别模型家族，不同 ch
 
 # 最新动态 🔥
 
-- **FunASR 1.4.14** 是当前 Python 发布版，覆盖源码安装、MOSS 发现与实时/工业部署。安装命令：`python -m pip install -U "funasr==1.4.14"`。[发布说明 ->](https://github.com/modelscope/FunASR/releases/tag/v1.4.14)
+- **FunASR 1.4.15** 是当前 Python 发布版，覆盖源码安装、MOSS 发现与实时/工业部署。安装命令：`python -m pip install -U "funasr==1.4.15"`。[发布说明 ->](https://github.com/modelscope/FunASR/releases/tag/v1.4.15)
 - **MOSS-Transcribe-Diarize** 是 OpenMOSS 的第三方模型，可离线完成长音频转写、时间戳和匿名说话人标签；FunASR 已提供服务、Docker、Kubernetes、vLLM、SGLang、LocalAI 与 FunClip 部署路径。[部署 MOSS ->](https://www.funasr.com/deploy/moss-transcribe-diarize.html)
 - **工业部署** 覆盖实时 WebSocket 服务、原生 vLLM 批量/流式路径，以及已校验的 Linux、macOS、Windows llama.cpp / GGUF 包。[Runtime v0.2.6 ->](https://github.com/modelscope/FunASR/releases/tag/runtime-llamacpp-v0.2.6) · [vLLM 指南 ->](docs/vllm_guide_zh.md)
 
+# Transformers 原生快速开始
+
+直接使用已发布的 Transformers 5.17.0，不需要安装 FunASR 工具库或执行模型仓库的远程 Python 代码。基础 Nano 支持中、英、日；31 语言 MLT 是另一个 checkpoint。
+
+```bash
+python -m pip install 'transformers==5.17.0' 'torch==2.10.0' 'torchaudio==2.10.0' 'librosa==0.11.0' 'soundfile==0.13.1'
+```
+
+[完整 Python 示例](https://www.funasr.com/docs/native-transformers.html) · [本地音频、批处理与热词](examples/transformers/) · [Notebook](examples/colab/fun_asr_nano_transformers.ipynb) · [Space](https://huggingface.co/spaces/FunAudioLLM/Fun-ASR-Nano)
+
+```python
+import torch
+from transformers import AutoModelForSpeechSeq2Seq, AutoProcessor
+
+torch.set_num_threads(4)
+model_id = "FunAudioLLM/Fun-ASR-Nano-2512-hf"
+revision = "d93b302ee7fd505e1b3576120fc142fc6f7820e1"
+audio = "https://huggingface.co/FunAudioLLM/Fun-ASR-Nano-2512/resolve/272c57b82523ada6fd87095e955f8e29100979ab/example/en.mp3"
+
+processor = AutoProcessor.from_pretrained(
+    model_id, revision=revision, trust_remote_code=False, token=False
+)
+model = AutoModelForSpeechSeq2Seq.from_pretrained(
+    model_id, revision=revision, trust_remote_code=False, token=False,
+    dtype=torch.float32,
+).to("cpu").eval()
+inputs = processor.apply_transcription_request(
+    audio=audio, language="en",
+    processor_kwargs={
+        "return_tensors": "pt",
+        "audio_kwargs": {"sampling_rate": 16000},
+        "text_kwargs": {"padding": True},
+    },
+)
+with torch.inference_mode():
+    generated = model.generate(**inputs, max_new_tokens=128, do_sample=False)
+new_tokens = generated[:, inputs.input_ids.shape[1]:]
+print(processor.batch_decode(new_tokens, skip_special_tokens=True)[0])
+```
+
 # 核心特性 🎯
 
 **Fun-ASR** 专注于高精度语音识别、checkpoint 级多语言支持和行业定制化能力。

examples/README.md

@@ -1,7 +1,16 @@
 # Runnable examples
 
-These scripts mirror the main README snippets and are intended to run from a
-fresh clone. Install the base requirements first:
+## Native Transformers (start here for the Hugging Face API)
+
+Use the [native quickstart](transformers/) for the released `transformers==5.17.0`
+package and the official `Fun-ASR-Nano-2512-hf` checkpoint. It includes a tested
+CPU script, local audio, batching, keywords and an [upload notebook](colab/fun_asr_nano_transformers.ipynb).
+This path does not require the FunASR toolkit or repository-local remote code.
+
+## FunASR toolkit examples
+
+The remaining scripts use the original toolkit checkpoint and a separate
+environment. Install the base requirements for these scripts only:
 
 ```bash
 pip install -r requirements.txt