t5-base-pointer-adv-mtop

This model is a fine-tuned version of google/mt5-base on the mtop dataset. It achieves the following results on the evaluation set:

• Loss: 0.1281
• Exact Match: 0.7105

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
1.7704	1.09	200	0.3664	0.1315
1.9751	2.17	400	0.2091	0.3400
1.0019	3.26	600	0.1453	0.4586
1.313	4.35	800	0.1313	0.5065
0.6593	5.43	1000	0.1281	0.5266
0.3216	6.52	1200	0.1317	0.5253
0.4614	7.61	1400	0.1508	0.5262
0.3577	8.69	1600	0.1422	0.5360
0.3748	9.78	1800	0.1419	0.5459
0.2422	10.87	2000	0.1603	0.5356
0.4443	11.96	2200	0.1526	0.5472
0.2671	13.04	2400	0.1606	0.5481
0.227	14.13	2600	0.1774	0.5441
0.2053	15.22	2800	0.1752	0.5441
0.1517	16.3	3000	0.1770	0.5481

Framework versions

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