llama-3-8b-base-orpo-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: 1.2904
• Rewards/chosen: -0.0901
• Rewards/rejected: -0.1085
• Rewards/accuracies: 0.6048
• Rewards/margins: 0.0184
• Logps/rejected: -1.0849
• Logps/chosen: -0.9013
• Logits/rejected: -0.4914
• Logits/chosen: -0.5121
• Nll Loss: 1.2288
• Log Odds Ratio: -0.6491
• Log Odds Chosen: 0.2841

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	Nll Loss	Log Odds Ratio	Log Odds Chosen
5.5193	0.4188	200	1.3328	-0.0905	-0.1087	0.5806	0.0183	-1.0874	-0.9048	-0.5090	-0.5265	1.2685	-0.6496	0.2829
5.2034	0.8377	400	1.2904	-0.0901	-0.1085	0.6048	0.0184	-1.0849	-0.9013	-0.4914	-0.5121	1.2288	-0.6491	0.2841

Framework versions

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