Update README with demo link, fix file names and sizes

README.md

@@ -10,83 +10,87 @@ tags:
 - streaming
 - q4_0
 - gguf
+- burn
 base_model: kyutai/stt-1b-en_fr
 ---
 
-  # STT 1B EN/FR — Q4 WebGPU
+# STT 1B EN/FR — Q4 WebGPU
 
-  Q4-quantized weights for [kyutai/stt-1b-en_fr](https://huggingface.co/kyutai/stt-1b-en_fr), packaged for client-side browser inference via WASM + WebGPU.
+Q4-quantized weights for [kyutai/stt-1b-en_fr](https://huggingface.co/kyutai/stt-1b-en_fr), packaged for client-side browser inference via WASM + WebGPU.
 
-  Runs entirely in the browser — no server required. English + French, streaming, ~1B parameters.
+Runs entirely in the browser — no server required. English + French, streaming, ~1B parameters.
 
-  ## Files
+[**Try the demo →**](https://idle-intelligence.github.io/stt-web/web/)
 
-  | File | Size | Description |
-  |------|------|-------------|
-  | `stt-1b-en_fr-q4.gguf` | 531 MB | STT transformer weights (Q4_0 quantized) |
-  | `mimi.safetensors` | 367 MB | Mimi audio codec weights (f32) |
-  | `tokenizer.model` | 118 KB | SentencePiece tokenizer (32k vocab, EN+FR) |
+## Files
 
-  ## Usage
+| File | Size | Description |
+|------|------|-------------|
+| `stt-1b-en_fr-q4_0.gguf` | 531 MB | STT transformer weights (Q4_0 quantized) |
+| `mimi-encoder-f16.safetensors` | 107 MB | Mimi audio codec encoder (f16) |
+| `tokenizer.model` | 118 KB | SentencePiece tokenizer (32k vocab, EN+FR) |
 
-  These weights are consumed by [stt-web](https://github.com/idle-intelligence/stt-web), a Rust/WASM + WebGPU speech-to-text engine.
+## Usage
 
-  ```js
-  import { SttClient } from './stt-client.js';
+These weights are consumed by [stt-web](https://github.com/idle-intelligence/stt-web), a Rust/WASM + WebGPU speech-to-text engine built with [Burn](https://burn.dev).
 
-  const stt = new SttClient({
-      baseUrl: 'https://your-site.com/stt',
-      shardList: ['stt-1b-en_fr-q4.gguf'],
-      mimiUrl: 'https://huggingface.co/idle-intelligence/stt-1b-en_fr-q4-webgpu/resolve/main/mimi.safetensors',
-      tokenizerUrl: 'https://huggingface.co/idle-intelligence/stt-1b-en_fr-q4-webgpu/resolve/main/tokenizer.model',
-      onTranscript: (text) => console.log(text),
-  });
+```js
+import { SttClient } from './stt-client.js';
 
-  await stt.init();
-  await stt.startRecording();
+const stt = new SttClient({
+    onTranscript: (text, isFinal) => console.log(text),
+    onStatus: (text, ready) => console.log(text),
+});
 
+await stt.init();
+await stt.startRecording();
 ```
 
-  Requirements
+Model weights are fetched from this repo automatically and cached by the browser.
 
-  - Chrome 113+ or Edge 113+ (WebGPU required)
-  - HTTPS (required for WebGPU)
-  - ~1 GB download on first load (cached by browser afterward)
+## Requirements
 
-  Pipeline
+- Chrome 113+ or Edge 113+ (WebGPU required)
+- HTTPS (required for WebGPU)
+- ~640 MB download on first load (cached afterward)
 
-  Microphone → AudioWorklet (24kHz mono)
-    → Mimi codec [WASM, CPU] → 32 codebook tokens/frame at 12.5Hz
-      → STT transformer [WASM, WebGPU] → text tokens
-        → SentencePiece detokenizer → transcript
+## Pipeline
 
-  Model Details
+```
+Microphone → AudioWorklet (24kHz mono)
+  → Mimi codec [WASM, CPU] → 32 codebook tokens/frame at 12.5Hz
+    → STT transformer [WASM, WebGPU] → text tokens
+      → SentencePiece detokenizer → transcript
+```
 
-  - Base model: https://huggingface.co/kyutai/stt-1b-en_fr by https://kyutai.org
-  - Architecture: Decoder-only transformer with delayed-streams modeling
-  - Parameters: ~1B (STT) + ~25M (Mimi codec)
-  - Quantization: Q4_0 (4-bit) for STT transformer, f32 for Mimi codec
-  - Languages: English, French
-  - Streaming latency: ~500ms text delay (6 frames at 12.5Hz)
-  - License: CC-BY 4.0 (same as original)
+## Model Details
 
-  Quantization
+- **Base model:** [kyutai/stt-1b-en_fr](https://huggingface.co/kyutai/stt-1b-en_fr) by [Kyutai](https://kyutai.org)
+- **Architecture:** Decoder-only transformer with delayed-streams modeling
+- **Parameters:** ~1B (STT) + ~25M (Mimi codec encoder)
+- **Quantization:** Q4_0 (4-bit) for STT transformer, f16 for Mimi codec
+- **Languages:** English, French
+- **Streaming latency:** ~500ms text delay (6 frames at 12.5Hz)
+- **License:** CC-BY 4.0 (same as original)
 
-  The STT transformer weights were quantized from f32 to Q4_0 using a custom GGUF packer. Dequantization happens on-GPU via WGSL compute shaders at inference time. The Mimi codec remains at f32 as it runs on
-   CPU and is small enough not to benefit from quantization.
+## Quantization
 
-  Citation
+The STT transformer weights were quantized from f32 to Q4_0 using a custom GGUF packer. Dequantization happens on-GPU via WGSL compute shaders at inference time. The Mimi codec encoder is stored at f16 as it runs on CPU via WASM.
 
-  If you use this model, please cite the original authors:
+## Citation
 
-  @techreport{kyutai2024stt,
-      author = {Kyutai},
-      title = {Speech-To-Text models},
-      institution = {Kyutai},
-      year = {2024},
-      url = {https://huggingface.co/kyutai/stt-1b-en_fr},
-  }
+If you use this model, please cite the original authors:
+
+```bibtex
+@techreport{kyutai2024stt,
+    author = {Kyutai},
+    title = {Speech-To-Text models},
+    institution = {Kyutai},
+    year = {2024},
+    url = {https://huggingface.co/kyutai/stt-1b-en_fr},
+}
+```
 
-  Disclaimer
+## Disclaimer
 
-  This is an independent port, not affiliated with or endorsed by Kyutai Labs. Transcription quality may differ from the original PyTorch implementation due to quantization.
+This is an independent port by [idle intelligence](https://idleintelligence.org), not affiliated with or endorsed by Kyutai Labs. Transcription quality may differ from the original PyTorch implementation due to quantization.