Pixie.30.32

This model is a fine-tuned version of t5-small on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.1623
• Bleu: 47.6437

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: 1e-05
• train_batch_size: 32
• eval_batch_size: 32
• seed: 121
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 30

Training results

Training Loss	Epoch	Step	Validation Loss	Bleu
No log	9.09	100	1.5563	21.3462
No log	18.18	200	1.2493	29.2353
No log	27.27	300	1.1670	32.5700

Framework versions

• Transformers 4.18.0
• Pytorch 1.11.0a0+17540c5
• Datasets 2.1.0
• Tokenizers 0.12.1