metadata
license: cc-by-4.0
base_model: HPLT/translate-mr-en-v2.0-hplt_opus
library_name: ctranslate2
pipeline_tag: translation
tags:
- translation
- marian
- ctranslate2
- int8
- mr
- en
translate-windy-hplt-mr-en — mr to en
Bilingual machine translation in CTranslate2 INT8 for fast CPU inference.
Fine-tuned by Windstorm Labs from HPLT/translate-mr-en-v2.0-hplt_opus.
Attribution
Derived from HPLT/translate-mr-en-v2.0-hplt_opus, from the HPLT project, licensed under
CC-BY-4.0. Modified by Windstorm Labs. Attribution is retained as the licence requires.
What we did
Converted from the upstream Marian release, LoRA fine-tuned on parallel data for this language pair, merged into the base weights, then quantized to INT8.
| Rank / alpha | 8 / 16 |
| Learning rate / steps | 2.5e-06 / 50 |
| Target modules | q_proj, v_proj |
| Seed | 42 |
| Training data | opus100_mr-en (clinic cache), 2,000 sentence pairs |
| Tensors modified | 36 |
Provenance
base model.bin sha256 7014cf14366ae8ef8bc83459feb265ed8bdfb71f8884741ca85cb79dc3ee1cf9
this model.bin sha256 4b424ffbbcf29b2514c8a1d60d6bacea9cfaa6c7341c4e451fc40987682b1734
Verified after INT8 quantization, on the published artifact.
Evaluation
FLORES-200 devtest, 200 sentences, beam 4, CTranslate2 INT8 on CPU.
| spBLEU | chrF | |
|---|---|---|
| this model | 28.39 | 55.13 |
Notes
- Single language pair: mr to en.
- FLORES-200 is news and encyclopedic prose.
- No human evaluation was performed.