---
license: apache-2.0
base_model: Helsinki-NLP/opus-mt-es-ar
library_name: transformers
pipeline_tag: translation
tags:
- translation
- marian
---

# translate-es-ar

Machine translation model, `es` to `ar`, published by Windstorm Labs.

## Attribution

Derived from [`Helsinki-NLP/opus-mt-es-ar`](https://huggingface.co/Helsinki-NLP/opus-mt-es-ar), licensed **apache-2.0**.
Windstorm Labs did not train the original model. This notice provides the attribution the
licence requires.

## What was changed

**The weights in this repository have been modified from the original.** A deliberate, documented and reproducible numerical transformation applied to the base weights. It is not a training run and is not described as one.

| | |
|---|---|
| Method | `epsilon-noise-fog` |
| Tensors modified | **72 of 254** |
| Max absolute weight delta | **5.234e-07** |
| Verified | tensor-by-tensor against the upstream original, 2026-07-26 |

The comparison is tensor-level rather than file-level: safetensors and PyTorch `.bin`
containers hash differently even when the tensors inside are identical, so a file-hash
mismatch would prove nothing.

## Contents

A single transformers build in safetensors format, loadable directly:

```python
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

tok = AutoTokenizer.from_pretrained("WindyTranslate/translate-es-ar")
model = AutoModelForSeq2SeqLM.from_pretrained("WindyTranslate/translate-es-ar")

batch = tok(["<your text here>"], return_tensors="pt", padding=True)
print(tok.batch_decode(model.generate(**batch), skip_special_tokens=True))
```

Other builds of this pair, including CTranslate2 INT8, are not included in this repository.

## Limitations

- A single language direction: `es` to `ar`.
- **No quality benchmark has been run on this pair.** The weights are verified to differ from
  the original; that is a statement about provenance, not about translation quality.
- Quality is not monotonic with model size or with the amount of fine-tuning applied.