bbc-to-ind-nmt-v7

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: 1.1540
• Sacrebleu: 38.1839
• Gen Len: 37.279

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: 16
• 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
5.0915	1.0	413	2.2849	27.9598	38.462
1.483	2.0	826	1.2052	35.2398	37.6305
1.0733	3.0	1239	1.1450	36.4283	37.133
0.9415	4.0	1652	1.1232	37.7264	37.198
0.8558	5.0	2065	1.1231	37.9682	37.399
0.7867	6.0	2478	1.1286	38.272	37.4305
0.736	7.0	2891	1.1343	38.0986	37.31
0.696	8.0	3304	1.1416	38.2159	37.219
0.6674	9.0	3717	1.1494	38.2257	37.307
0.6488	10.0	4130	1.1540	38.1839	37.279

Framework versions

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