Add model card: attribution, licence, and change statement
Browse files
README.md
ADDED
|
@@ -0,0 +1,57 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
base_model: Helsinki-NLP/opus-mt-cpp-cpp
|
| 4 |
+
library_name: transformers
|
| 5 |
+
pipeline_tag: translation
|
| 6 |
+
tags:
|
| 7 |
+
- translation
|
| 8 |
+
- marian
|
| 9 |
+
---
|
| 10 |
+
|
| 11 |
+
# translate-cpp-cpp
|
| 12 |
+
|
| 13 |
+
Machine translation model, `cpp` to `cpp`, published by Windstorm Labs.
|
| 14 |
+
|
| 15 |
+
## Attribution
|
| 16 |
+
|
| 17 |
+
Derived from [`Helsinki-NLP/opus-mt-cpp-cpp`](https://huggingface.co/Helsinki-NLP/opus-mt-cpp-cpp), licensed **apache-2.0**.
|
| 18 |
+
Windstorm Labs did not train the original model. This notice provides the attribution the
|
| 19 |
+
licence requires.
|
| 20 |
+
|
| 21 |
+
## What was changed
|
| 22 |
+
|
| 23 |
+
**The weights in this repository have been modified from the original.** A LoRA fine-tune trained on parallel corpus data and merged into the base weights.
|
| 24 |
+
|
| 25 |
+
| | |
|
| 26 |
+
|---|---|
|
| 27 |
+
| Method | `lora-fog` |
|
| 28 |
+
| Tensors modified | **36 of 254** |
|
| 29 |
+
| Max absolute weight delta | **6.841e-05** |
|
| 30 |
+
| Verified | tensor-by-tensor against the upstream original, 2026-07-26 |
|
| 31 |
+
|
| 32 |
+
The comparison is tensor-level rather than file-level: safetensors and PyTorch `.bin`
|
| 33 |
+
containers hash differently even when the tensors inside are identical, so a file-hash
|
| 34 |
+
mismatch would prove nothing.
|
| 35 |
+
|
| 36 |
+
## Contents
|
| 37 |
+
|
| 38 |
+
A single transformers build in safetensors format, loadable directly:
|
| 39 |
+
|
| 40 |
+
```python
|
| 41 |
+
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
|
| 42 |
+
|
| 43 |
+
tok = AutoTokenizer.from_pretrained("WindyTranslate/translate-cpp-cpp")
|
| 44 |
+
model = AutoModelForSeq2SeqLM.from_pretrained("WindyTranslate/translate-cpp-cpp")
|
| 45 |
+
|
| 46 |
+
batch = tok(["<your text here>"], return_tensors="pt", padding=True)
|
| 47 |
+
print(tok.batch_decode(model.generate(**batch), skip_special_tokens=True))
|
| 48 |
+
```
|
| 49 |
+
|
| 50 |
+
Other builds of this pair, including CTranslate2 INT8, are not included in this repository.
|
| 51 |
+
|
| 52 |
+
## Limitations
|
| 53 |
+
|
| 54 |
+
- A single language direction: `cpp` to `cpp`.
|
| 55 |
+
- **No quality benchmark has been run on this pair.** The weights are verified to differ from
|
| 56 |
+
the original; that is a statement about provenance, not about translation quality.
|
| 57 |
+
- Quality is not monotonic with model size or with the amount of fine-tuning applied.
|