---
library_name: peft
license: mit
base_model: facebook/m2m100_418M
tags:
- base_model:adapter:facebook/m2m100_418M
- lora
- transformers
model-index:
- name: m2m100_418M-GEC-spanish-LORA-cowsl2h
  results: []
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# m2m100_418M-GEC-spanish-LORA-cowsl2h

This model is a fine-tuned version of [facebook/m2m100_418M](https://huggingface.co/facebook/m2m100_418M) on an unknown dataset.
It achieves the following results on the evaluation set:
- Loss: 6.0424
- Gleu: 0.6151

## Model description

More information needed

## Intended uses & limitations

More information needed

## Training and evaluation data

More information needed

## Training procedure

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 0.00044107500561578277
- train_batch_size: 16
- eval_batch_size: 16
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- optimizer: Use OptimizerNames.ADAFACTOR and the args are:
No additional optimizer arguments
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 0.08950396527674138
- num_epochs: 4
- mixed_precision_training: Native AMP

### Training results

| Training Loss | Epoch | Step | Validation Loss | Gleu   |
|:-------------:|:-----:|:----:|:---------------:|:------:|
| 14.5713       | 1.0   | 194  | 6.0505          | 0.5815 |
| 12.1501       | 2.0   | 388  | 6.0441          | 0.6024 |
| 12.1355       | 3.0   | 582  | 6.0436          | 0.6056 |
| 12.1091       | 4.0   | 776  | 6.0424          | 0.6151 |


### Framework versions

- PEFT 0.18.1
- Transformers 5.0.0
- Pytorch 2.7.0+cu126
- Datasets 4.8.5
- Tokenizers 0.22.2