bbc-to-ind-nmt-v6

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

• Loss: 1.1940
• Sacrebleu: 38.2052
• Gen Len: 37.467

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: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Sacrebleu	Gen Len
4.1507	1.0	825	1.3826	31.3365	37.6035
1.2319	2.0	1650	1.1646	36.0321	37.553
0.9893	3.0	2475	1.1238	37.2804	37.284
0.8593	4.0	3300	1.1213	38.1118	37.409
0.7624	5.0	4125	1.1353	38.2863	37.234
0.6872	6.0	4950	1.1404	38.3932	37.1405
0.6253	7.0	5775	1.1566	38.1803	37.191
0.5781	8.0	6600	1.1723	38.3633	37.351
0.5441	9.0	7425	1.1836	38.25	37.485
0.5214	10.0	8250	1.1940	38.2052	37.467

Framework versions

• Transformers 4.41.2
• Pytorch 2.3.0+cu121
• Datasets 2.14.6
• Tokenizers 0.19.1