Add Quick Start section to README with prebuilt binary instructions

Signed-off-by: Michael Yuan <michael@secondstate.io>
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: juntao

README.md

@@ -6,121 +6,108 @@ Pure Rust implementation of [Qwen3-ASR](https://github.com/QwenLM/Qwen3-ASR) aut
 
 Supports two backends: **libtorch** (via the `tch` crate, cross-platform with optional CUDA) and **MLX** (Apple Silicon native via Metal GPU). Loads model weights directly from safetensors files and re-implements the complete neural network forward pass in Rust.
 
-## Architecture
+## Quick Start
 
-The implementation ports the Qwen3-ASR encoder-decoder architecture from PyTorch/Transformers to Rust with libtorch (via the `tch` crate):
+### 1. Download the binary
 
-- **Audio Encoder** (Whisper-style): 3x Conv2d downsampling → sinusoidal positional embeddings → 18 transformer encoder layers → output projection (896 → 1024)
-- **Text Decoder** (Qwen3): 28 transformer decoder layers with Grouped Query Attention (16 Q heads / 8 KV heads), QK-normalization, MRoPE (Multimodal Rotary Position Embeddings), and SwiGLU MLP
-- **Audio preprocessing**: FFmpeg decodes any audio format → resampled to mono 16kHz f32 → 128-bin log-mel spectrogram (Whisper-style)
+Download the latest release for your platform from [GitHub Releases](https://github.com/second-state/qwen3_asr_rs/releases/latest) and extract:
 
-## Supported Models
+**macOS (Apple Silicon)**
 
-| Model | Parameters | HuggingFace |
-|-------|-----------|-------------|
-| Qwen3-ASR-0.6B | 0.6B | [Qwen/Qwen3-ASR-0.6B](https://huggingface.co/Qwen/Qwen3-ASR-0.6B) |
-| Qwen3-ASR-1.7B | 1.7B | [Qwen/Qwen3-ASR-1.7B](https://huggingface.co/Qwen/Qwen3-ASR-1.7B) |
+```bash
+curl -LO https://github.com/second-state/qwen3_asr_rs/releases/latest/download/asr-macos-aarch64.zip
+unzip asr-macos-aarch64.zip
+# Contains: asr-macos-aarch64/asr and asr-macos-aarch64/mlx.metallib
+```
 
-## Prerequisites
+**Linux x86_64 (CPU)**
 
-### Backend
+```bash
+curl -LO https://github.com/second-state/qwen3_asr_rs/releases/latest/download/asr-linux-x86_64.zip
+unzip asr-linux-x86_64.zip
+# Contains: asr-linux-x86_64/asr
+```
 
-Choose one backend:
+**Linux x86_64 (CUDA)**
 
-| Backend | Feature flag | Platforms | GPU |
-|---------|-------------|-----------|-----|
-| libtorch | `tch-backend` (default) | Linux, macOS, Windows | CUDA |
-| MLX | `mlx` | macOS Apple Silicon | Metal |
-
-### libtorch (for `tch-backend`)
+```bash
+curl -LO https://github.com/second-state/qwen3_asr_rs/releases/latest/download/asr-linux-x86_64-cuda.zip
+unzip asr-linux-x86_64-cuda.zip
+# Contains: asr-linux-x86_64-cuda/asr
+```
 
-The `tch` crate (v0.20) requires **libtorch 2.7.1**. Download and extract for your platform:
+**Linux ARM64**
 
 ```bash
-# macOS (Apple Silicon)
-curl -LO https://download.pytorch.org/libtorch/cpu/libtorch-macos-arm64-2.7.1.zip
-unzip libtorch-macos-arm64-2.7.1.zip
+curl -LO https://github.com/second-state/qwen3_asr_rs/releases/latest/download/asr-linux-aarch64.zip
+unzip asr-linux-aarch64.zip
+# Contains: asr-linux-aarch64/asr
+```
+
+### 2. Download libtorch (Linux only)
 
+macOS uses the MLX backend and does not need libtorch.
+
+```bash
 # Linux x86_64 (CPU)
 curl -LO https://download.pytorch.org/libtorch/cpu/libtorch-cxx11-abi-shared-with-deps-2.7.1%2Bcpu.zip
 unzip libtorch-cxx11-abi-shared-with-deps-2.7.1+cpu.zip
 
-# Linux ARM64 (CPU)
-curl -LO https://github.com/second-state/libtorch-releases/releases/download/v2.7.1/libtorch-cxx11-abi-aarch64-2.7.1.tar.gz
-tar xzf libtorch-cxx11-abi-aarch64-2.7.1.tar.gz
-
 # Linux x86_64 (CUDA 12.8)
 curl -LO https://download.pytorch.org/libtorch/cu128/libtorch-cxx11-abi-shared-with-deps-2.7.1%2Bcu128.zip
 unzip libtorch-cxx11-abi-shared-with-deps-2.7.1+cu128.zip
-```
 
-### FFmpeg
-
-Install FFmpeg development libraries:
-
-```bash
-# macOS
-brew install ffmpeg
-
-# Ubuntu/Debian
-sudo apt-get install libavcodec-dev libavformat-dev libavutil-dev libswresample-dev pkg-config
+# Linux ARM64
+curl -LO https://github.com/second-state/libtorch-releases/releases/download/v2.7.1/libtorch-cxx11-abi-aarch64-2.7.1.tar.gz
+tar xzf libtorch-cxx11-abi-aarch64-2.7.1.tar.gz
 ```
 
-### Model Weights
-
-Download a model from HuggingFace:
+### 3. Download model weights
 
 ```bash
-# 0.6B model
-huggingface-cli download Qwen/Qwen3-ASR-0.6B --local-dir Qwen3-ASR-0.6B
+pip install huggingface_hub transformers
 
-# 1.7B model (sharded safetensors, auto-detected)
-huggingface-cli download Qwen/Qwen3-ASR-1.7B --local-dir Qwen3-ASR-1.7B
-```
-
-Generate `tokenizer.json` (required):
+huggingface-cli download Qwen/Qwen3-ASR-0.6B --local-dir Qwen3-ASR-0.6B
 
-```bash
 python -c "
 from transformers import AutoTokenizer
 tok = AutoTokenizer.from_pretrained('Qwen3-ASR-0.6B', trust_remote_code=True)
 tok.backend_tokenizer.save('Qwen3-ASR-0.6B/tokenizer.json')
 "
 ```
 
-## Build
-
-### libtorch backend (default)
+### 4. Transcribe
 
 ```bash
-# Set environment
-export LIBTORCH=$(pwd)/libtorch
-export LIBTORCH_BYPASS_VERSION_CHECK=1
-export LD_LIBRARY_PATH=$LIBTORCH/lib:$LD_LIBRARY_PATH    # Linux
-export DYLD_LIBRARY_PATH=$LIBTORCH/lib:$DYLD_LIBRARY_PATH  # macOS
+# macOS (MLX backend — no extra env needed)
+./asr-macos-aarch64/asr Qwen3-ASR-0.6B input.wav
 
-# Build (dynamically links FFmpeg)
-cargo build --release
+# Linux
+LD_LIBRARY_PATH=$(pwd)/libtorch/lib:$LD_LIBRARY_PATH \
+  ./asr-linux-x86_64/asr Qwen3-ASR-0.6B input.wav
+```
 
-# Build with statically linked FFmpeg
-cargo build --release --features static-ffmpeg
+Output:
 
-# Build FFmpeg from source and link statically (most self-contained)
-cargo build --release --features build-ffmpeg
+```
+Language: English
+Text: Thank you for your contribution to the most recent issue of Computer.
 ```
 
-### MLX backend (macOS Apple Silicon)
+## Architecture
 
-```bash
-# Initialize mlx-c submodule
-git submodule update --init --recursive
+The implementation ports the Qwen3-ASR encoder-decoder architecture from PyTorch/Transformers to Rust with libtorch (via the `tch` crate):
 
-# Build with MLX (no libtorch needed)
-cargo build --release --no-default-features --features mlx
+- **Audio Encoder** (Whisper-style): 3x Conv2d downsampling → sinusoidal positional embeddings → 18 transformer encoder layers → output projection (896 → 1024)
+- **Text Decoder** (Qwen3): 28 transformer decoder layers with Grouped Query Attention (16 Q heads / 8 KV heads), QK-normalization, MRoPE (Multimodal Rotary Position Embeddings), and SwiGLU MLP
+- **Audio preprocessing**: FFmpeg decodes any audio format → resampled to mono 16kHz f32 → 128-bin log-mel spectrogram (Whisper-style)
 
-# With statically linked FFmpeg
-cargo build --release --no-default-features --features mlx,static-ffmpeg
-```
+## Supported Models
+
+| Model | Parameters | HuggingFace |
+|-------|-----------|-------------|
+| Qwen3-ASR-0.6B | 0.6B | [Qwen/Qwen3-ASR-0.6B](https://huggingface.co/Qwen/Qwen3-ASR-0.6B) |
+| Qwen3-ASR-1.7B | 1.7B | [Qwen/Qwen3-ASR-1.7B](https://huggingface.co/Qwen/Qwen3-ASR-1.7B) |
 
 ## Usage
 
@@ -132,19 +119,10 @@ asr ./Qwen3-ASR-0.6B input.wav
 asr ./Qwen3-ASR-0.6B input.wav chinese
 asr ./Qwen3-ASR-0.6B input.wav english
 
-# Any audio format (FFmpeg handles conversion)
-asr ./Qwen3-ASR-0.6B input.mp3
-asr ./Qwen3-ASR-0.6B input.flac
-asr ./Qwen3-ASR-0.6B input.m4a
-
 # Enable debug logging
 RUST_LOG=debug asr ./Qwen3-ASR-0.6B input.wav
 ```
 
-### Input Audio Requirements
-
-The `asr` binary accepts **any audio format** supported by FFmpeg. The audio is automatically converted to mono 16 kHz f32 for the model's mel spectrogram computation.
-
 ### Output Format
 
 ```
@@ -156,6 +134,63 @@ Text: 你好世界
 
 Qwen3-ASR supports 30 languages: Chinese, English, Cantonese, Arabic, German, French, Spanish, Portuguese, Indonesian, Italian, Korean, Russian, Thai, Vietnamese, Japanese, Turkish, Hindi, Malay, Dutch, Swedish, Danish, Finnish, Polish, Czech, Filipino, Persian, Greek, Romanian, Hungarian, Macedonian.
 
+## Build from Source
+
+### Backend
+
+Choose one backend:
+
+| Backend | Feature flag | Platforms | GPU |
+|---------|-------------|-----------|-----|
+| libtorch | `tch-backend` (default) | Linux, macOS, Windows | CUDA |
+| MLX | `mlx` | macOS Apple Silicon | Metal |
+
+### Prerequisites
+
+**libtorch** (for `tch-backend`): See [Step 2](#2-download-libtorch-linux-only) above for download links.
+
+**FFmpeg** development libraries:
+
+```bash
+# macOS
+brew install ffmpeg
+
+# Ubuntu/Debian
+sudo apt-get install libavcodec-dev libavformat-dev libavutil-dev libswresample-dev pkg-config
+```
+
+### libtorch backend (default)
+
+```bash
+# Set environment
+export LIBTORCH=$(pwd)/libtorch
+export LIBTORCH_BYPASS_VERSION_CHECK=1
+export LD_LIBRARY_PATH=$LIBTORCH/lib:$LD_LIBRARY_PATH    # Linux
+export DYLD_LIBRARY_PATH=$LIBTORCH/lib:$DYLD_LIBRARY_PATH  # macOS
+
+# Build (dynamically links FFmpeg)
+cargo build --release
+
+# Build with statically linked FFmpeg
+cargo build --release --features static-ffmpeg
+
+# Build FFmpeg from source and link statically (most self-contained)
+cargo build --release --features build-ffmpeg
+```
+
+### MLX backend (macOS Apple Silicon)
+
+```bash
+# Initialize mlx-c submodule
+git submodule update --init --recursive
+
+# Build with MLX (no libtorch needed)
+cargo build --release --no-default-features --features mlx
+
+# With statically linked FFmpeg
+cargo build --release --no-default-features --features mlx,static-ffmpeg
+```
+
 ## Project Structure
 
 ```