llama-3-8b-base-ipo-ultrafeedback-8xh200

This model is a fine-tuned version of W-61/llama-3-8b-base-sft-ultrachat-8xh200 on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 20.1926
• Rewards/chosen: -0.1172
• Rewards/rejected: -0.2002
• Rewards/accuracies: 0.7621
• Rewards/margins: 0.0830
• Logps/rejected: -3.3356
• Logps/chosen: -2.3107
• Logits/rejected: -0.6593
• Logits/chosen: -0.6771

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-07
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• total_eval_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
87.5502	0.4188	200	21.8803	-0.0519	-0.1012	0.7056	0.0493	-2.3452	-1.6582	-0.7170	-0.7410
80.9952	0.8377	400	20.1926	-0.1172	-0.2002	0.7621	0.0830	-3.3356	-2.3107	-0.6593	-0.6771

Framework versions

• Transformers 4.51.0
• Pytorch 2.3.1+cu121
• Datasets 2.21.0
• Tokenizers 0.21.4