facebook-NLLB-arb-fr

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: 0.8957
• Bleu: 36.4987
• Rouge: 0.4885
• Meteor: 0.4899
• Gen Len: 31.9385

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: 64
• eval_batch_size: 64
• 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
• num_epochs: 50
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Rouge	Meteor	Gen Len
2.0158	1.0331	500	1.3732	18.1484	0.3203	0.3303	32.0566
1.4316	2.0661	1000	1.1694	24.7767	0.3809	0.3899	31.7446
1.2478	3.0992	1500	1.0773	27.7293	0.4079	0.4154	31.9393
1.1356	4.1322	2000	1.0231	29.7626	0.4273	0.4327	31.7922
1.0558	5.1653	2500	0.9883	31.3369	0.4408	0.4459	31.9325
0.9948	6.1983	3000	0.9643	32.4082	0.4506	0.4547	31.8666
0.9418	7.2314	3500	0.9449	33.1697	0.4573	0.4619	31.977
0.9024	8.2645	4000	0.9302	33.7465	0.4617	0.467	31.8995
0.8657	9.2975	4500	0.9189	33.9253	0.4641	0.4672	31.6999
0.8316	10.3306	5000	0.9158	34.4241	0.4691	0.4706	31.504
0.8006	11.3636	5500	0.9061	35.0012	0.4735	0.4757	31.8126
0.7789	12.3967	6000	0.9053	35.3364	0.4745	0.4789	31.9359
0.7523	13.4298	6500	0.9002	35.599	0.4776	0.4804	31.6772
0.7343	14.4628	7000	0.8966	35.7566	0.4811	0.4835	31.8183
0.7084	15.4959	7500	0.9000	35.7263	0.4814	0.4838	31.7661
0.6909	16.5289	8000	0.8939	36.1187	0.4841	0.487	31.8682
0.6743	17.5620	8500	0.8974	36.3555	0.4883	0.4894	31.7302
0.6559	18.5950	9000	0.8966	36.3941	0.4891	0.491	31.7578
0.6413	19.6281	9500	0.8957	36.4987	0.4885	0.4899	31.9385

Framework versions

• Transformers 5.0.0
• Pytorch 2.10.0+cu128
• Datasets 4.0.0
• Tokenizers 0.22.2