hi-en / README.md
AmerAlreyahi's picture
Add OptimaMT hi-en
28a576b verified
|
Raw History Blame Contribute Delete
2.6 kB
---
license: cc-by-4.0
language:
- hi
- en
pipeline_tag: translation
library_name: transformers.js
base_model: Xenova/opus-mt-hi-en
tags:
- translation
- marian
- opus-mt
- onnx
---
# OptimaMT hi → en (Hindi to English)
Drop-in copy of [Xenova/opus-mt-hi-en](https://huggingface.co/Xenova/opus-mt-hi-en) at revision `99853100ed9d` with a lighter merged decoder: ONNX Runtime's fast mode (`disable_quant_qdq = 1`, 2.3–2.5× faster) at half the memory. Same weights, same tokens.
Part of [OptimaMT](https://github.com/AmerAlreyahi/OptimaMT): [all 51 pairs](https://huggingface.co/collections/OptimaMT/optimamt-lighter-opus-mt-onnx-models-6af228dc9cd91f015e56e869), named `OptimaMT/<src>-<tgt>`.
**Used in production:**
- [OpenCircle](https://github.com/AmerAlreyahi/OpenCircle), a desktop screen translator for Windows and Linux, ships these models since v0.1.4.
| Memory after load | Linux | Windows | Identical tokens |
|---|---:|---:|---|
| `disable_quant_qdq = 1` | 518 → 278 MiB (-46 %) | 491 → 256 MiB (-48 %) | yes |
| ONNX Runtime defaults | 277 → 217 MiB (-22 %) | 194 → 164 MiB (-15 %) | yes |
FLORES-200 devtest (1,012 sentences): the same translations as the original, BLEU 18.11, chrF 44.76 (greedy).
ONNX Runtime 1.31.1, CPU, 4 threads (Linux; Windows in a VM). Interactive results: [OptimaMT/results](https://huggingface.co/spaces/OptimaMT/results) ·
method: [REPORT.md](https://github.com/AmerAlreyahi/OptimaMT/blob/main/REPORT.md).
## Use
```js
import { pipeline } from "@huggingface/transformers";
const translator = await pipeline("translation", "OptimaMT/hi-en", {
dtype: "q8",
session_options: { extra: { session: { disable_quant_qdq: "1" } } },
});
const [out] = await translator("प्रकाशस्तंभ तट पर खड़ा है।");
console.log(out.translation_text);
```
Python ([translate.py](https://github.com/AmerAlreyahi/OptimaMT/blob/main/scripts/translate.py): onnxruntime, sentencepiece, numpy):
```bash
python translate.py OptimaMT/hi-en "प्रकाशस्तंभ तट पर खड़ा है।"
```
## Files
Only `onnx/decoder_model_merged_quantized.onnx` is rebuilt; every other file is the original, byte for byte.
Vocabulary: 61,127 tokens.
## Licence and credits
- Weights: [OPUS-MT](https://huggingface.co/Helsinki-NLP/opus-mt-hi-en) by the University of Helsinki, CC BY 4.0.
- ONNX export: [Xenova/opus-mt-hi-en](https://huggingface.co/Xenova/opus-mt-hi-en) at revision `99853100ed9d`.
- Modified: the decoder's output projection was moved out of its two branches (graph change only). CC BY 4.0.