t5-small-pointer-cstop_artificial

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

• Loss: 0.0816
• Exact Match: 0.8050

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: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 32
• 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
2.08	28.5	200	0.3320	0.0376
0.272	57.13	400	0.1084	0.2630
0.0789	85.63	600	0.0830	0.3184
0.0355	114.25	800	0.0816	0.3363
0.0207	142.75	1000	0.0868	0.3292
0.014	171.38	1200	0.0952	0.3399
0.0099	199.88	1400	0.1089	0.3381
0.0076	228.5	1600	0.1104	0.3381
0.0057	257.13	1800	0.1153	0.3292
0.0048	285.63	2000	0.1153	0.3327
0.004	314.25	2200	0.1206	0.3363
0.0032	342.75	2400	0.1229	0.3363
0.0028	371.38	2600	0.1268	0.3381
0.0023	399.88	2800	0.1288	0.3399
0.002	428.5	3000	0.1292	0.3399

Framework versions

• Transformers 4.25.1
• Pytorch 1.13.0+cu117
• Datasets 2.7.1
• Tokenizers 0.13.2