byt5-base-mtop

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

• Loss: 0.0704
• Exact Match: 0.7978

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
0.5459	6.65	200	0.0964	0.2107
0.0433	13.33	400	0.0704	0.3056
0.0153	19.98	600	0.0764	0.3025
0.0081	26.65	800	0.0815	0.3123
0.0045	33.33	1000	0.0904	0.3065
0.003	39.98	1200	0.0945	0.3060
0.0021	46.65	1400	0.0969	0.3087
0.0016	53.33	1600	0.1017	0.3105
0.0015	59.98	1800	0.1001	0.3096
0.001	66.65	2000	0.1059	0.3087
0.0006	73.33	2200	0.1041	0.3105
0.0005	79.98	2400	0.1083	0.3083
0.0003	86.65	2600	0.1115	0.3096
0.0002	93.33	2800	0.1130	0.3119
0.0002	99.98	3000	0.1131	0.3114

Framework versions

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