Pin validated inference dependencies and refresh setup guides

README.md

@@ -38,7 +38,7 @@ English · [简体中文](https://huggingface.co/inclusionAI/Ming-3.0-asr-tiny/b
 
 ### Download and install
 
-Prepare compatible CUDA builds of PyTorch and torchaudio, then download the complete checkpoint and install its remaining dependencies:
+Use Python 3.12 on a Linux machine with a compatible CUDA GPU. Download the complete checkpoint and install the tested dependencies, including PyTorch and torchaudio:
 
 ```bash
 python -m pip install "huggingface_hub[cli]"
@@ -56,7 +56,7 @@ python examples/transcribe.py \
   --audio-chunk-frames 6000
 ```
 
-Add `--language-hint zh` or `--language-hint en` when the language is known. The example uses CUDA, BF16, SDPA, and greedy decoding; it loads the bundled model code with `trust_remote_code=True`. The reference environment used Transformers 5.12.1, fla-core 0.5.2, and PyTorch 2.13.0+cu130.
+Add `--language-hint zh` or `--language-hint en` when the language is known. The example uses CUDA, BF16, SDPA, and greedy decoding; it loads the bundled model code with `trust_remote_code=True`. The tested environment used Transformers 5.12.1, fla-core 0.5.2, and PyTorch 2.11.0+cu130. For another CUDA version, first install matching PyTorch and torchaudio 2.11.0 wheels using the [official PyTorch instructions](https://pytorch.org/get-started/previous-versions/), then run `pip install -r requirements.txt`.
 
 ### PyTorch / Transformers Python API
 

README.zh.md

@@ -25,7 +25,7 @@
 
 ### 下载与安装
 
-先准备兼容 CUDA 的 PyTorch 和 torchaudio，再下载完整模型并安装其余依赖：
+使用 Python 3.12 和兼容 CUDA 的 Linux GPU。下载完整模型后，按已验证的依赖版本安装，包括 PyTorch 与 torchaudio：
 
 ```bash
 python -m pip install "huggingface_hub[cli]"
@@ -43,7 +43,7 @@ python examples/transcribe.py \
   --audio-chunk-frames 6000
 ```
 
-已知语种时可加 `--language-hint zh` 或 `--language-hint en`。示例使用 CUDA、BF16、SDPA 和 greedy decoding，并通过 `trust_remote_code=True` 加载仓库内的自定义实现。参考验证环境为 Transformers 5.12.1、fla-core 0.5.2 和 PyTorch 2.13.0+cu130。
+已知语种时可加 `--language-hint zh` 或 `--language-hint en`。示例使用 CUDA、BF16、SDPA 和 greedy decoding，并通过 `trust_remote_code=True` 加载仓库内的自定义实现。已验证环境为 Transformers 5.12.1、fla-core 0.5.2 和 PyTorch 2.11.0+cu130。若需其他 CUDA 版本，先按 [PyTorch 官方说明](https://pytorch.org/get-started/previous-versions/)安装匹配的 PyTorch 与 torchaudio 2.11.0 wheel，再运行 `pip install -r requirements.txt`。
 
 ### PyTorch / Transformers Python 示例
 

requirements.txt

@@ -1,9 +1,11 @@
-# Runtime interface validated with Transformers 5.12.1 and fla-core 0.5.2.
-# Install mutually compatible CUDA builds of torch and torchaudio first.
-transformers>=5.12.1,<5.13
+# Validated on Python 3.12, Linux, CUDA 13.0 (NVIDIA L20Y).
+# For another CUDA version, install matching PyTorch 2.11.0 wheels first.
+torch==2.11.0
+torchaudio==2.11.0
+transformers==5.12.1
 fla-core==0.5.2
-accelerate
-einops
-safetensors
-numpy
-soundfile
+accelerate==1.15.0
+einops==0.8.2
+safetensors==0.8.0
+numpy==2.5.4
+soundfile==0.14.0