m2m100_418m-eng-nso-onnx

m2m100_418m-eng-nso-onnx

ONNX export of dsfsi/m2m100_418m-eng-nso.

Translation direction: English (en) -> Northern Sotho / Sepedi (nso).

Attribution

The original model was trained by DSFSI (Data Science for Social Impact research group, University of Pretoria), a fine-tune of facebook/m2m100_418M. Licence: Apache-2.0, as declared by the original repository. This repository only converts the weights to ONNX. All credit for the model belongs to DSFSI.

Export

optimum-cli export onnx --model dsfsi/m2m100_418m-eng-nso \
  --task text2text-generation-with-past --no-post-process <outdir>

The int8 quantized files needed the .onnx.data -> .onnx_data rename and proto location rewrite (encoder, decoder and decoder-with-past, plus one Constant node tensor attribute per file). Verified with a real load from a cache cleared beforehand.

Files

Parity

10 general-domain English sentences, PyTorch original vs ONNX, exact string match against AutoModelForSeq2SeqLM.generate().

parity:
  sample_size: 10
  metric: exact_match
  fp32_greedy: 1.00   # 10/10
  fp32_beam4:  1.00   # 10/10
  int8_greedy: 0.60   # 6/10
  int8_beam4:  0.70   # 7/10

fp32 is an exact reproduction of the original model. int8 disagreements were inspected by hand and are paraphrase-level Northern Sotho variation (word choice, pronoun form), not garbling or wrong-language output.

Caveat on sample size: n=10 exact-match is a small, noisy estimate, not a FLORES-scale benchmark. A separate measurement campaign on this project found n=5 exact-match numbers for another model swing from 100% to 14% when re-measured at n=100 on FLORES-200. Treat the percentages above as an initial signal only - a proper FLORES + chrF evaluation supersedes it wherever available.

Selecting the language

Northern Sotho's M2M100 language code is ns (not nso). The target is not baked into forced_bos_token_id in generation_config.json for this checkpoint - you must set it explicitly:

tokenizer.src_lang = "en"
forced_bos_token_id = tokenizer.lang_code_to_id["ns"]
model.generate(**inputs, forced_bos_token_id=forced_bos_token_id, ...)

Usage

from transformers import AutoTokenizer
from optimum.onnxruntime import ORTModelForSeq2SeqLM

repo = "TigreGotico/m2m100_418m-eng-nso-onnx"
tok = AutoTokenizer.from_pretrained(repo)
tok.src_lang = "en"
model = ORTModelForSeq2SeqLM.from_pretrained(repo, use_cache=True, use_merged=False)

enc = tok("The weather is very nice today.", return_tensors="pt")
tgt_id = tok.lang_code_to_id["ns"]
out = model.generate(**enc, forced_bos_token_id=tgt_id, num_beams=4, max_new_tokens=64)
print(tok.batch_decode(out, skip_special_tokens=True)[0])
# -> Boso bo le bonolo kudu mo lebakeng le.

For the int8 build, pass subfolder="int8".