File size: 1,662 Bytes
2e74133
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
---
license: cc-by-4.0
base_model: HPLT/translate-en-vi-v2.0-hplt_opus
library_name: ctranslate2
pipeline_tag: translation
tags: [translation, marian, ctranslate2, int8, en, vi]
---

# translate-windy-hplt-en — en to vi

Bilingual machine translation in **CTranslate2 INT8** for fast CPU inference.
Fine-tuned by **Windstorm Labs** from [`HPLT/translate-en-vi-v2.0-hplt_opus`](https://huggingface.co/HPLT/translate-en-vi-v2.0-hplt_opus).

## Attribution

Derived from [`HPLT/translate-en-vi-v2.0-hplt_opus`](https://huggingface.co/HPLT/translate-en-vi-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_en-vi (clinic cache), 2,000 sentence pairs |
| Tensors modified | 36 |

## Provenance

```
base model.bin  sha256  0c94dc9fcf527ddab4e9df98dee176b401213aa3fbaf9c4feed5623e439b9c99
this model.bin  sha256  d2108b1d5cf28a080d3ca20b17a17f313ed6f731a24a9bb9e155a0c05d08b1a9
```

Verified after INT8 quantization, on the published artifact.

## Evaluation

FLORES-200 devtest, 200 sentences, beam 4, CTranslate2 INT8 on CPU.

| | spBLEU | chrF |
|---|---:|---:|
| this model | **40.35** | 57.7 |

## Notes

- Single language pair: en to vi.
- FLORES-200 is news and encyclopedic prose.
- No human evaluation was performed.