Update model card for int8 quantized models

README.md

@@ -11,12 +11,14 @@ tags:
   - onnx
   - transformers.js
   - txukun
+  - quantized
+  - int8
 pipeline_tag: translation
 ---
 
 # Txukun — Capitalization & Punctuation Restoration (ONNX)
 
-ONNX export of [HiTZ/cap-punct-eu](https://huggingface.co/HiTZ/cap-punct-eu) for browser and edge inference with Transformers.js.
+ONNX export of [HiTZ/cap-punct-eu](https://huggingface.co/HiTZ/cap-punct-eu) for browser and edge inference with Transformers.js. Weights are dynamically quantized to int8 for faster downloads (77 MB vs 297 MB fp32).
 
 ## Model
 
@@ -29,7 +31,7 @@ MarianMT model trained on 9.78M Basque sentences that restores capitalization an
 - **Type**: MarianMT (6 encoder + 6 decoder layers)
 - **d_model**: 512
 - **Tokenizer**: SentencePiece (Unigram, 32k vocab), shipped as custom `tokenizer.json`
-- **Format**: Float32 ONNX (required by ORT Web WASM — no float16 support on CPU backend)
+- **Format**: Int8 dynamically quantized ONNX (Q/DQ nodes, compatible with ORT Web WASM)
 - **Input**: Lowercase, punctuationless Basque text
 - **Output**: Properly capitalized and punctuated Basque text
 
@@ -37,8 +39,8 @@ MarianMT model trained on 9.78M Basque sentences that restores capitalization an
 
 | File | Size | Description |
 |------|------|-------------|
-| `encoder_model.onnx` | 136 MB | Encoder (IR 8, opset 18, fp32) |
-| `decoder_model_merged.onnx` | 161 MB | Decoder with KV-cache (IR 8, opset 18, fp32) |
+| `encoder_model_quantized.onnx` | 34 MB | Encoder (IR 8, int8 quantized) |
+| `decoder_model_merged_quantized.onnx` | 41 MB | Decoder with KV-cache (IR 8, int8 quantized) |
 | `tokenizer.json` | 2.1 MB | Custom Unigram + Metaspace pre-tokenizer |
 | `config.json` | 979 B | Model configuration |
 | `tokenizer_config.json` | 864 B | Tokenizer metadata |
@@ -52,7 +54,7 @@ import { pipeline } from '@huggingface/transformers';
 const corrector = await pipeline(
   'translation',
   'itzune/txukun-cap-punct-eu',
-  { device: 'wasm' }
+  { device: 'wasm', dtype: 'q8' }
 );
 
 const result = await corrector('kaixo zer moduz zaude');
@@ -60,6 +62,14 @@ console.log(result[0].translation_text);
 // → "Kaixo, zer moduz zaude?"
 ```
 
+## Quantization details
+
+Dynamically quantized with `onnxruntime.quantization.quantize_dynamic(QuantType.QInt8, extra_options={"EnableSubgraph": True})`. The `EnableSubgraph` flag traverses into the `If`-node subgraphs of the merged decoder, quantizing `MatMul` operations in both branches. Results:
+
+- Encoder: 136 MB → 34 MB (75% reduction)
+- Decoder: 160 MB → 41 MB (74% reduction)
+- **Total: 297 MB → 77 MB (74% reduction)**
+
 ## Part of Txukun
 
 This model powers [Txukun](https://github.com/itzune/txukun), a browser-based Basque text cleaning tool. Visit [itzune.eus/txukun](https://itzune.eus/txukun/) to try it.