opus-mt-fi-cs ONNX int8 (movil offline)
Conversion movil-ready de Helsinki-NLP/opus-mt-fi-cs.
ONNX
- encoder_model_quantized.onnx
- decoder_model_quantized.onnx
- decoder_with_past_model_quantized.onnx (KV cache, decoding O(1) por token)
Quant: int8 dynamic ARM64 + per_channel.
Tokenizer
- tokenizer.json (Fast / Xenova / transformers.js)
- source.spm, target.spm, vocab.json (Marian raw)
Source: Helsinki-NLP/opus-mt-fi-cs
Repo: R4kSo1997/opus-mt-fi-cs-onnx-int8