t5-small-finetuned-en-to-ro-lr_2e-3-fp_false

This model is a fine-tuned version of t5-small on the wmt16 dataset. It achieves the following results on the evaluation set:

• Loss: 1.4239
• Bleu: 7.1921
• Gen Len: 18.2611

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: 16
• eval_batch_size: 16
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu	Gen Len
0.8922	0.05	2000	1.7000	6.5274	18.2656
0.8621	0.1	4000	1.6409	6.6411	18.2311
0.8433	0.16	6000	1.6396	6.6601	18.2596
0.8297	0.21	8000	1.6304	6.7129	18.2581
0.8006	0.26	10000	1.6022	6.6067	18.2816
0.793	0.31	12000	1.5999	6.551	18.2631
0.774	0.37	14000	1.5586	6.7105	18.2661
0.7618	0.42	16000	1.5769	6.7278	18.2526
0.7463	0.47	18000	1.5625	6.6972	18.2201
0.7394	0.52	20000	1.5377	6.936	18.2491
0.7203	0.58	22000	1.5191	7.0205	18.2731
0.7158	0.63	24000	1.5055	6.835	18.2506
0.688	0.68	26000	1.4779	7.0534	18.2716
0.678	0.73	28000	1.4691	6.9735	18.2616
0.6677	0.79	30000	1.4702	7.0359	18.2496
0.6568	0.84	32000	1.4534	6.9982	18.2556
0.6475	0.89	34000	1.4427	7.0443	18.2466
0.6395	0.94	36000	1.4265	7.1205	18.2721
0.6319	1.0	38000	1.4239	7.1921	18.2611

Framework versions

• Transformers 4.12.5
• Pytorch 1.10.0+cu111
• Datasets 1.16.1
• Tokenizers 0.10.3