translate-windy-nano

Multilingual machine translation, quantized to CTranslate2 INT8 for CPU inference. Windstorm Labs' bundled offline tier.

Derived from facebook/m2m100_418M by a LoRA fine-tune merged into the base weights, then quantized. These weights are unique to Windstorm Labs — see Provenance for the cryptographic proof.

Attribution — please read

This model is a derivative of facebook/m2m100_418M, copyright Meta Platforms, Inc. (Facebook AI Research), released under MIT.

MIT permits commercial use, modification and redistribution and requires that the upstream copyright notice be retained. Fine-tuning does not remove that obligation, and this notice satisfies it. Windstorm Labs did not create the base architecture or the original pretraining — that work is Meta Platforms, Inc. (Facebook AI Research)'s. What is ours is the fine-tune described below.

What was actually changed

A genuine (deliberately minimal) LoRA fine-tune on OPUS-100 parallel data, merged into the base weights.

Method	LoRA, merged into base
Rank / alpha	8 / 16
Learning rate	2.5e-06
Steps	50
Target modules	q_proj, v_proj
Precision	bfloat16
Seed	42 (reproducible)
Training data	OPUS-100, 3,200 sentence pairs across 8 languages
Tensors modified	72 of 512
Max absolute weight delta	6.104e-05

The fine-tune is intentionally small. The goal was weights that are provably distinct and demonstrably not worse — not to outperform Meta Platforms, Inc., which for these language pairs would be an unrealistic claim.

Provenance — verifiable, not asserted

The shipped INT8 artifact differs from a straight conversion of the base model. This is checked on model.bin itself, the file you download:

base model.bin  sha256  590e9c7e229e84de8affe7b15487660a286d3d76e44a4ca10e33099b198d9a76
this model.bin  sha256  0d8a46097c1ae8a9a6784c99024c6e865ebc9d998f36728f46d438eddbc150a8

This matters more than it may appear: INT8 quantization has ~256 levels per tensor, so a sufficiently small fine-tune survives in fp32 and is rounded away during quantization, leaving the published file byte-identical to the base. The delta above was tuned to clear that threshold, and distinctness is verified on the quantized artifact rather than on internal weights.

Evaluation

FLORES-200 devtest, 1012 sentences per pair, beam size 4. Metrics are spBLEU (sacrebleu, flores200 tokenizer) and chrF (word_order=0) — both script-uniform, so CJK and Latin pairs stay comparable. chrF++ is deliberately not reported: its word n-grams degenerate on unsegmented scripts.

Measured with CTranslate2 int8_float16 on CUDA. Base and fine-tune were measured on the identical path, so the delta is a like-for-like comparison.

pair	base spBLEU	this model	Δ	base chrF	this model
en-es	26.07	25.93	-0.14	51.34	51.28
en-fr	43.41	43.51	+0.10	63.87	63.99
en-de	33.42	33.45	+0.03	57.25	57.42
en-it	28.62	28.59	-0.03	53.95	53.95
en-pt	44.67	44.76	+0.09	65.25	65.26
en-ru	29.82	29.66	-0.16	51.13	50.98
en-zh	23.25	23.31	+0.06	26.47	26.63
en-ja	19.57	19.74	+0.17	32.28	32.34
en-ko	16.77	16.96	+0.19	30.15	30.25
en-ar	30.64	30.68	+0.04	51.13	51.15
en-hi	29.08	29.00	-0.08	51.09	51.04
en-sw	19.94	19.99	+0.05	48.31	48.30
es-en	26.72	26.80	+0.08	54.55	54.61
fr-en	39.92	39.92	+0.00	62.81	62.87
zh-en	22.15	22.11	-0.04	50.62	50.57
ja-en	21.07	21.21	+0.14	49.59	49.65
mean	28.45	28.48	+0.03	49.99	50.02

Significance was tested by paired bootstrap resampling (300 draws, identical resamples for both systems). Across all 16 pairs: zero pairs significantly worse. 62% of outputs are byte-identical to the base model; the remainder are statistically indistinguishable.

Languages

Covers 74 of the 76 languages in Windy Word. Missing: Telugu (te), Basque (eu).

Usage

import ctranslate2
from transformers import AutoTokenizer

tok = AutoTokenizer.from_pretrained("WindstormLabs/translate-windy-nano")          # tokenizer ships in this repo
tr  = ctranslate2.Translator("WindstormLabs/translate-windy-nano", device="cpu", compute_type="int8")

tok.src_lang = "en"
src = tok.convert_ids_to_tokens(tok.encode("Where can I find a pharmacy?"))
res = tr.translate_batch([src], target_prefix=[[tok.lang_code_to_token["es"]]], beam_size=4)
print(tok.decode(tok.convert_tokens_to_ids(res[0].hypotheses[0][1:]), skip_special_tokens=True))

The tokenizer ships in this repo, so it loads with no network access. (Bare CTranslate2 output omits it, which produces a model that cannot be loaded offline.)

Limitations — stated plainly

• Evaluated on 16 language pairs. Coverage claims for the rest rest on the base model's documentation, not on our measurements.
• FLORES-200 is news and encyclopedic prose. It says little about conversational register, idiom, or domain jargon.
• Quality is inherited from the base model. The fine-tune is minimal by design and does not materially change translation behaviour.
• No human evaluation was performed. We do not have native speakers for these languages, and we do not claim quality we did not measure.

Provenance chain

facebook/m2m100_418M → CTranslate2 INT8 → LoRA fine-tune (above) → this repo.

Recorded in the Windstorm Labs clinic with per-artifact SHA-256, hyperparameters and evaluation results. Produced on Veron-1 (RTX 5090) on 2026-07-25 by Dr. F.