opus-mt-id-en-finetuned-id-to-en

This model is a fine-tuned version of Helsinki-NLP/opus-mt-id-en on the id_panl_bppt dataset. It achieves the following results on the evaluation set:

• Loss: 1.6469
• Bleu: 30.557
• Gen Len: 29.8247

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-06
• train_batch_size: 32
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Gen Len
2.5737	1.0	751	2.2222	24.4223	30.3344
2.3756	2.0	1502	2.1264	25.419	30.3147
2.3146	3.0	2253	2.0588	26.0995	30.1959
2.2411	4.0	3004	2.0072	26.5944	30.0763
2.1927	5.0	3755	1.9657	27.0422	30.0773
2.1554	6.0	4506	1.9284	27.4151	30.0715
2.1105	7.0	5257	1.8980	27.6645	29.9426
2.0841	8.0	6008	1.8680	28.023	29.9797
2.0491	9.0	6759	1.8438	28.2456	29.9342
2.0265	10.0	7510	1.8218	28.5378	29.8968
2.0065	11.0	8261	1.8012	28.7599	29.8907
1.9764	12.0	9012	1.7835	28.9369	29.8796
1.969	13.0	9763	1.7663	29.1565	29.8671
1.9474	14.0	10514	1.7506	29.3313	29.893
1.9397	15.0	11265	1.7378	29.4567	29.8512
1.9217	16.0	12016	1.7239	29.6245	29.8361
1.9174	17.0	12767	1.7127	29.7464	29.8398
1.9021	18.0	13518	1.7030	29.9035	29.8621
1.89	19.0	14269	1.6934	29.9669	29.8225
1.878	20.0	15020	1.6847	30.0961	29.8398
1.8671	21.0	15771	1.6774	30.1878	29.839
1.8634	22.0	16522	1.6717	30.2341	29.8134
1.8536	23.0	17273	1.6653	30.3356	29.816
1.8533	24.0	18024	1.6602	30.3548	29.8251
1.8476	25.0	18775	1.6560	30.4323	29.8315
1.8362	26.0	19526	1.6528	30.4682	29.8277
1.8463	27.0	20277	1.6501	30.5002	29.8236
1.8369	28.0	21028	1.6484	30.5236	29.8257
1.8313	29.0	21779	1.6472	30.55	29.8259
1.8332	30.0	22530	1.6469	30.557	29.8247

Framework versions

• Transformers 4.21.1
• Pytorch 1.12.0
• Datasets 2.4.0
• Tokenizers 0.12.1