bbc-to-ind-nmt-v5

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: 1.2310
• Sacrebleu: 38.4814
• Gen Len: 37.8455

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: 4
• 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
3.4154	1.0	1650	1.2829	33.2857	37.9245
1.1633	2.0	3300	1.1418	36.6342	37.407
0.9377	3.0	4950	1.1148	38.0023	37.17
0.795	4.0	6600	1.1197	38.2402	37.3695
0.6827	5.0	8250	1.1465	38.3719	37.315
0.5937	6.0	9900	1.1642	38.3424	37.547
0.5216	7.0	11550	1.1917	38.56	37.8515
0.466	8.0	13200	1.2079	38.6061	37.6135
0.425	9.0	14850	1.2228	38.4918	37.928
0.3995	10.0	16500	1.2310	38.4814	37.8455

Framework versions

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