nllb_600m-en-swh-drL-24_5-filtered-v24_28_2

This model is a fine-tuned version of facebook/nllb-200-distilled-600M on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4517
• Bleu Eng Latn-swh Latn: 75.6335
• Bleu Swh Latn-eng Latn: 61.5722

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu Eng Latn-swh Latn	Bleu Swh Latn-eng Latn
0.5402	1.0	2439	0.4881	71.3066	59.7916
0.4466	2.0	4878	0.4628	73.6631	60.7412
0.3875	3.0	7317	0.4529	74.8640	61.0596
0.3515	4.0	9756	0.4503	75.3634	61.4414
0.3254	5.0	12195	0.4517	75.6335	61.5722

Framework versions

• Transformers 4.43.3
• Pytorch 2.3.1+cu121
• Datasets 2.14.7
• Tokenizers 0.19.1