dalio-large-test

This model is a fine-tuned version of facebook/opt-125m on the AlekseyKorshuk/dalio-handwritten-io dataset. It achieves the following results on the evaluation set:

• Loss: 3.1016
• Accuracy: 0.0477

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: 3e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 8
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• training_steps: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
3.3342	0.05	1	3.1016	0.0477

Framework versions

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