PontifexMaximus/Turkish2

opus-mt-tr-en-finetuned-tr-to-en

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

• Loss: 0.6321
• Bleu: 56.617
• Gen Len: 13.5983

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: 64
• eval_batch_size: 64
• 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
No log	1.0	241	1.2487	41.0053	13.0461
No log	2.0	482	1.1630	43.1077	13.0386
1.4091	3.0	723	1.0992	44.6583	13.0445
1.4091	4.0	964	1.0463	45.5931	13.0289
1.2325	5.0	1205	1.0012	46.7039	12.9998
1.2325	6.0	1446	0.9610	47.6783	13.0274
1.1284	7.0	1687	0.9262	48.622	12.9866
1.1284	8.0	1928	0.8939	48.4984	13.5762
1.0486	9.0	2169	0.8642	49.1496	13.5918
1.0486	10.0	2410	0.8391	49.8875	13.5905
0.9866	11.0	2651	0.8150	50.6447	13.5803
0.9866	12.0	2892	0.7941	51.2059	13.5731
0.9362	13.0	3133	0.7741	51.7071	13.5754
0.9362	14.0	3374	0.7564	52.4185	13.5781
0.8928	15.0	3615	0.7398	53.0814	13.5744
0.8928	16.0	3856	0.7247	53.5711	13.5783
0.8598	17.0	4097	0.7111	54.0559	13.568
0.8598	18.0	4338	0.6988	54.5188	13.5598
0.8274	19.0	4579	0.6876	54.78	13.5765
0.8274	20.0	4820	0.6780	55.1494	13.5762
0.8086	21.0	5061	0.6688	55.5813	13.5788
0.8086	22.0	5302	0.6610	55.6403	13.5796
0.7878	23.0	5543	0.6539	55.7731	13.5989
0.7878	24.0	5784	0.6483	55.9956	13.593
0.7718	25.0	6025	0.6432	56.2303	13.5904
0.7718	26.0	6266	0.6390	56.4825	13.5975
0.7633	27.0	6507	0.6360	56.5334	13.5958
0.7633	28.0	6748	0.6338	56.5357	13.5965
0.7633	29.0	6989	0.6325	56.5862	13.5974
0.7584	30.0	7230	0.6321	56.617	13.5983

Framework versions

• Transformers 4.20.1
• Pytorch 1.12.0
• Datasets 2.3.2
• Tokenizers 0.12.1