opus-mt-itc-itc-onnx

opus-mt-itc-itc-onnx

ONNX export of Helsinki-NLP/opus-mt-itc-itc, a MarianMT multilingual translation model covering Italic languages (Italian, Catalan, Romansh, Spanish, Romanian, Galician, Sardinian, Corsican, Walloon, Portuguese, Occitan, Aragonese, Indonesian, French, Haitian Creole, and more) translating between each other.

Exported for use with optimum.onnxruntime.ORTModelForSeq2SeqLM / onnxruntime, e.g. in phoonnx / OVOS translation routing.

Licence: apache-2.0 (verbatim from the upstream model card).

Attribution: model trained and published by Helsinki-NLP / University of Helsinki as part of the OPUS-MT project (Tatoeba-Challenge, itc-itc, opus-2020-07-07). This repository only republishes an ONNX conversion of their weights; no model changes were made.

Why this model

This model unblocks direct Portuguese → Spanish translation (opus-mt-pt-es / opus-mt-es-pt are both gone from the Helsinki-NLP org), avoiding a pivot through English for pt→eu style routing.

CRITICAL: target-language token required

This is a multilingual, many-to-many model. The target language is selected by a >>xxx<< token prepended to the input text — transformers/optimum does not add this automatically. Omitting it does not raise an error: the model silently produces fluent text in an unpredictable language.

Example: >>spa<< O tempo está bom hoje. → El tiempo es bueno hoy. (Portuguese → Spanish)

Full list of valid target tokens

>>arg<< >>ast<< >>bjn<< >>cat<< >>cos<< >>egl<< >>fra<< >>frm_Latn<< >>gcf_Latn<< >>glg<< >>hat<< >>ind<<
>>ita<< >>lad<< >>lad_Latn<< >>lat_Grek<< >>lat_Latn<< >>lij<< >>lld_Latn<< >>lmo<< >>mwl<< >>oci<< >>pap<<
>>pcd<< >>pms<< >>por<< >>roh<< >>ron<< >>scn<< >>spa<< >>srd<< >>vec<< >>wln<< >>zsm_Latn<<

(source: upstream model card language(s) list — arg ast bjn cat cos egl fra frm_Latn gcf_Latn glg hat ind ita lad lad_Latn lat_Grek lat_Latn lij lld_Latn lmo mwl oci pap pcd pms por roh ron scn spa srd vec wln zsm_Latn)

File layout

encoder_model.onnx, decoder_model.onnx, decoder_with_past_model.onnx      (fp32, repo root)
source.spm, target.spm, vocab.json, tokenizer_config.json, special_tokens_map.json, config.json, generation_config.json
int8/
  encoder_model.onnx, decoder_model.onnx, decoder_with_past_model.onnx    (dynamic QUInt8 quantization)
  (+ copies of the tokenizer files)

Export command

optimum-cli export onnx --model Helsinki-NLP/opus-mt-itc-itc \
  --task text2text-generation-with-past --no-post-process ./fp32

int8: onnxruntime.quantization.quantize_dynamic(weight_type=QUInt8) applied to each of the three graphs independently, written to fp32/int8/.

Parity

8 sentences covering pt→es, es→pt, ca→es, pt→ca, each with the correct >>xxx<< token, PyTorch (MarianMTModel) vs ONNX (ORTModelForSeq2SeqLM), num_beams=4, max_new_tokens=64, exact string match after decoding:

• fp32: 8/8 = 100%
• int8: 8/8 = 100%

Usage

from transformers import AutoTokenizer
from optimum.onnxruntime import ORTModelForSeq2SeqLM

tok = AutoTokenizer.from_pretrained("TigreGotico/opus-mt-itc-itc-onnx")
model = ORTModelForSeq2SeqLM.from_pretrained("TigreGotico/opus-mt-itc-itc-onnx")

text = ">>spa<< O tempo está bom hoje."
ids = model.generate(**tok(text, return_tensors="pt"), num_beams=4, max_new_tokens=64)
print(tok.decode(ids[0], skip_special_tokens=True))
# El tiempo es bueno hoy.

For int8, pass subfolder="int8" instead.