Instructions to use OptimaMT/en-it with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers.js
How to use OptimaMT/en-it with Transformers.js:
// npm i @huggingface/transformers import { pipeline } from '@huggingface/transformers'; // Allocate pipeline const pipe = await pipeline('translation', 'OptimaMT/en-it');
Add OptimaMT en-it
Browse files
README.md
CHANGED
|
@@ -15,14 +15,19 @@ tags:
|
|
| 15 |
|
| 16 |
# OptimaMT en → it (English to Italian)
|
| 17 |
|
| 18 |
-
Drop-in copy of [Xenova/opus-mt-en-it](https://huggingface.co/Xenova/opus-mt-en-it) at revision `075406e3c8c2` with a lighter merged decoder:
|
| 19 |
-
Part of [OptimaMT](https://github.com/AmerAlreyahi/OptimaMT)
|
|
|
|
|
|
|
|
|
|
| 20 |
|
| 21 |
| Memory after load | Linux | Windows | Identical tokens |
|
| 22 |
|---|---:|---:|---|
|
| 23 |
| `disable_quant_qdq = 1` | 645 → 333 MiB (-48 %) | 621 → 307 MiB (-51 %) | yes |
|
| 24 |
| ONNX Runtime defaults | 255 → 216 MiB (-15 %) | 230 → 190 MiB (-17 %) | yes |
|
| 25 |
|
|
|
|
|
|
|
| 26 |
ONNX Runtime 1.31.1, CPU, 4 threads (Linux; Windows in a VM). Interactive results: [OptimaMT/results](https://huggingface.co/spaces/OptimaMT/results) ·
|
| 27 |
method: [REPORT.md](https://github.com/AmerAlreyahi/OptimaMT/blob/main/REPORT.md).
|
| 28 |
|
|
|
|
| 15 |
|
| 16 |
# OptimaMT en → it (English to Italian)
|
| 17 |
|
| 18 |
+
Drop-in copy of [Xenova/opus-mt-en-it](https://huggingface.co/Xenova/opus-mt-en-it) at revision `075406e3c8c2` 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.
|
| 19 |
+
Part of [OptimaMT](https://github.com/AmerAlreyahi/OptimaMT): [all 49 pairs](https://huggingface.co/collections/OptimaMT/optimamt-lighter-opus-mt-onnx-models-6af228dc9cd91f015e56e869), named `OptimaMT/<src>-<tgt>`.
|
| 20 |
+
|
| 21 |
+
**Used in production:**
|
| 22 |
+
- [OpenCircle](https://github.com/AmerAlreyahi/OpenCircle), a desktop screen translator for Windows and Linux, ships these models since v0.1.4.
|
| 23 |
|
| 24 |
| Memory after load | Linux | Windows | Identical tokens |
|
| 25 |
|---|---:|---:|---|
|
| 26 |
| `disable_quant_qdq = 1` | 645 → 333 MiB (-48 %) | 621 → 307 MiB (-51 %) | yes |
|
| 27 |
| ONNX Runtime defaults | 255 → 216 MiB (-15 %) | 230 → 190 MiB (-17 %) | yes |
|
| 28 |
|
| 29 |
+
FLORES-200 devtest (1,012 sentences): the same translations as the original, BLEU 27.32, chrF 57.45 (greedy).
|
| 30 |
+
|
| 31 |
ONNX Runtime 1.31.1, CPU, 4 threads (Linux; Windows in a VM). Interactive results: [OptimaMT/results](https://huggingface.co/spaces/OptimaMT/results) ·
|
| 32 |
method: [REPORT.md](https://github.com/AmerAlreyahi/OptimaMT/blob/main/REPORT.md).
|
| 33 |
|