baseline.europarl

This model is a fine-tuned version of samzirbo/mT5.en-es.pretrained on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 3.5556
• Bleu: 28.4134
• Meteor: 0.548
• Chrf++: 50.7588

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: 0.0005
• train_batch_size: 64
• eval_batch_size: 64
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 1000
• training_steps: 50000

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Meteor	Chrf++
3.7368	0.06	2500	3.7330	16.6348	0.4263	38.9851
1.9501	0.11	5000	3.6988	20.5694	0.4632	42.6268
1.7374	0.17	7500	3.6554	21.8374	0.4825	44.4332
1.6252	0.23	10000	3.6674	23.1212	0.4954	45.8093
1.5421	0.28	12500	3.6407	24.144	0.5063	47.0076
1.4867	0.34	15000	3.6991	23.9884	0.5055	46.9973
1.4413	0.39	17500	3.6669	25.0356	0.5135	47.6982
1.4017	0.45	20000	3.5988	25.4766	0.5201	48.3754
1.3769	0.51	22500	3.6120	26.17	0.5295	49.037
1.3476	0.56	25000	3.6225	26.8343	0.5341	49.5501
1.3252	0.62	27500	3.5913	26.7117	0.5321	49.3981
1.307	0.68	30000	3.6205	27.3269	0.5385	49.9517
1.2926	0.73	32500	3.5624	27.7597	0.5446	50.3568
1.2823	0.79	35000	3.5449	27.8457	0.5458	50.5179
1.2728	0.85	37500	3.5383	27.9605	0.5444	50.3421
1.2663	0.9	40000	3.5556	28.1962	0.5465	50.5512
1.2628	0.96	42500	3.5498	28.3077	0.5477	50.7074
1.2514	1.01	45000	3.5519	28.3543	0.548	50.7669
1.239	1.07	47500	3.5555	28.2177	0.5473	50.6922
1.2357	1.13	50000	3.5556	28.4134	0.548	50.7588

Framework versions

• Transformers 4.38.0
• Pytorch 2.2.1+cu121
• Datasets 2.19.1
• Tokenizers 0.15.2