byt5-base-top_v2

This model is a fine-tuned version of google/byt5-base on the top_v2 dataset. It achieves the following results on the evaluation set:

• Loss: 0.0154
• Exact Match: 0.8533

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: 0.001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 64
• total_train_batch_size: 512
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• training_steps: 3000

Training results

Training Loss	Epoch	Step	Validation Loss	Exact Match
0.3866	0.82	200	0.0434	0.0278
0.0335	1.65	400	0.0224	0.0396
0.0206	2.47	600	0.0184	0.0403
0.0162	3.29	800	0.0189	0.0409
0.0165	4.12	1000	0.0167	0.0410
0.0116	4.94	1200	0.0154	0.0413
0.0093	5.76	1400	0.0154	0.0415
0.008	6.58	1600	0.0161	0.0413
0.0071	7.41	1800	0.0167	0.0414
0.0064	8.23	2000	0.0171	0.0414
0.0055	9.05	2200	0.0168	0.0415
0.0046	9.88	2400	0.0177	0.0415
0.0039	10.7	2600	0.0183	0.0416
0.0034	11.52	2800	0.0190	0.0414
0.0029	12.35	3000	0.0192	0.0415

Framework versions

• Transformers 4.24.0
• Pytorch 1.13.0+cu117
• Datasets 2.7.0
• Tokenizers 0.13.2