pythia-1.4b-deduped-jokes

This model is a fine-tuned version of EleutherAI/pythia-1.4b-deduped on the Fraser/short-jokes dataset. It achieves the following results on the evaluation set:

• Loss: 0.0699
• Accuracy: 0.9870

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: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• total_eval_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.05
• training_steps: 400

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
No log	0.06	100	0.0729	0.9866
No log	0.12	200	0.0716	0.9868
No log	0.17	300	0.0705	0.9869
No log	0.23	400	0.0699	0.9870

Framework versions

• Transformers 4.29.0.dev0
• Pytorch 2.0.0-rc1
• Datasets 2.11.0
• Tokenizers 0.13.3