mimba/nllb-malagasy2fra

This model is a fine-tuned version of facebook/nllb-200-distilled-600M on the mimba/text2text dataset. It achieves the following results on the evaluation set:

• Loss: 1.0624
• Bleu: 28.4830
• Chrf: 50.7145
• Wer: 0.6800

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: 2e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 688
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Chrf	Wer
1.5100	1.0	1722	1.1443	24.7796	47.4321	0.7152
1.1718	2.0	3444	1.0965	26.0499	48.4929	0.6989
1.0806	3.0	5166	1.0752	26.8260	49.3082	0.6952
1.0283	4.0	6888	1.0646	27.5526	49.7685	0.6829
0.9603	5.0	8610	1.0576	27.7081	50.0489	0.6861
0.9143	6.0	10332	1.0547	27.9679	50.2583	0.6835
0.8833	7.0	12054	1.0546	28.0900	50.3442	0.6797
0.8478	8.0	13776	1.0575	28.2149	50.4982	0.6792
0.8225	9.0	15498	1.0576	28.3610	50.5933	0.6788
0.8007	10.0	17220	1.0624	28.4830	50.7145	0.6800

Framework versions

• Transformers 5.0.0
• Pytorch 2.10.0+cu128
• Datasets 4.5.0
• Tokenizers 0.22.2