Commit History
-
docs: add Python/optimum usage example, list source.spm and vocab.json in files table
Verified signature2e20dc7 -
Upload tokenizer_config.json with huggingface_hub
Verified signaturebfc3ef9 -
Upload vocab.json with huggingface_hub
Verified signature602dd93 -
Upload source.spm with huggingface_hub
Verified signature873888d -
Update model card for int8 quantized models
Xabi Ezpeleta4e679d3 -
Add int8 quantized models (EnableSubgraph: True)
Xabi Ezpeleta- encoder_model_quantized.onnx: 136 MB -> 34 MB (75% reduction) - decoder_model_merged_quantized.onnx: 160 MB -> 41 MB (74% reduction) - Total: 297 MB -> 77 MB (74% reduction) Quantized with onnxruntime.quantization.quantize_dynamic + EnableSubgraph=True to traverse the If-node subgraphs in the merged decoder.d144555 -
Update model card for float32 (reverted from float16)
Xabi Ezpeleta1541b94 -
Revert to float32 — ONNX Runtime Web WASM does not support float16 tensors
Xabi EzpeletaFloat16 models produce Cast nodes that ORT Web WASM cannot execute. Reverting to the original float32 weights (297 MB total). An int8 quantization path exists for the encoder (34 MB, 75% reduction) but the decoder cannot be quantized due to the If-node wrapping from optimum's merged model export. Future work: extract subgraphs and quantize, or re-export decoder without If merge.dc7ef21 -
Rename ONNX files with _fp16 suffix for Transformers.js auto-detection
Xabi Ezpeleta1ec675c -
Update model card for float16
Xabi Ezpeleta09083af -
Quantize to float16 for smaller downloads
Xabi Ezpeleta- encoder_model.onnx: 136 MB -> 68 MB (50% reduction) - decoder_model_merged.onnx: 160 MB -> 81 MB (50% reduction) - Total: 297 MB -> 149 MB Converted with onnxconverter-common float16, IR version 8 preserved.84b30d7 -
Add model card
Xabi Ezpeleta1c28369 -
Initial ONNX model export from HiTZ/cap-punct-eu
Xabi Ezpeleta- encoder_model.onnx (IR 8, opset 18, 136MB) - decoder_model_merged.onnx (IR 8, opset 18, 161MB, KV-cache support) - tokenizer.json (custom-built Unigram with Metaspace pre-tokenizer) - Config files from original MarianMT model Exported via optimum-cli with seq2seq-lm-with-past, IR downgraded from 9 to 8 for ONNX Runtime Web WASM compatibility.34383e5 -
initial commit
Verified signaturec2386b1
