amazon-reviews-input-output

This model is a fine-tuned version of facebook/opt-350m on the AlekseyKorshuk/amazon-reviews-input-output dataset. It achieves the following results on the evaluation set:

• Loss: 3.0486
• Accuracy: 0.0373
• Samples: 100
• Perplexity: 21.0867
• Table: <wandb.data_types.Table object at 0x7fa288d36880>

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: 5e-05
• train_batch_size: 4
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 1.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
3.2766	0.8	25	3.0587	0.0372

Framework versions

• Transformers 4.25.0.dev0
• Pytorch 1.12.1+cu113
• Datasets 2.3.2
• Tokenizers 0.12.1