opus-mt-en-fr_wmt14_En_Fr_1million_20epochs_v2

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

• Loss: 6.7777
• Bleu: 0.0052
• Gen Len: 511.0

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: 0.002
• train_batch_size: 60
• eval_batch_size: 60
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Gen Len
2.0731	1.0	1667	4.9161	0.0	511.0
2.0281	2.0	3334	4.0396	0.0001	511.0
2.0226	3.0	5001	6.5526	0.0	511.0
1.9991	4.0	6668	4.6473	0.0001	511.0
1.991	5.0	8335	3.9736	0.0052	511.0
1.9749	6.0	10002	4.4897	0.0052	511.0
1.983	7.0	11669	5.7620	0.0052	511.0
1.9791	8.0	13336	5.9320	0.0	511.0
1.9494	9.0	15003	3.9506	0.0052	511.0
1.9482	10.0	16670	5.2391	0.0052	511.0
1.9328	11.0	18337	3.7053	0.0052	511.0
1.9201	12.0	20004	5.6524	0.0052	511.0
1.9242	13.0	21671	4.2519	0.0	511.0
1.9132	14.0	23338	6.6592	0.0052	511.0
1.9154	15.0	25005	6.1263	0.0052	511.0
1.9052	16.0	26672	7.1135	0.0052	511.0
1.9128	17.0	28339	6.1183	0.0	511.0
1.9015	18.0	30006	7.0262	0.0052	511.0
1.9025	19.0	31673	6.0193	0.0052	511.0
1.9015	20.0	33340	6.7777	0.0052	511.0

Framework versions

• Transformers 4.32.1
• Pytorch 1.12.1
• Datasets 2.18.0
• Tokenizers 0.13.2