Llama0-3-8b-ultra-p-0.05-lr1e-6-e1

This model is a fine-tuned version of meta-llama/Meta-Llama-3-8B-Instruct on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5155
• Rewards/chosen: -0.8052
• Rewards/rejected: -1.6601
• Rewards/accuracies: 0.75
• Rewards/margins: 0.8549
• Logps/rejected: -430.6040
• Logps/chosen: -337.1520
• Logits/rejected: 0.5600
• Logits/chosen: 0.4444

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: 1e-06
• train_batch_size: 2
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 8
• total_train_batch_size: 128
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.5887	0.2060	100	0.5813	-0.4118	-0.7745	0.6875	0.3628	-342.0493	-297.8121	0.1168	0.0514
0.5536	0.4119	200	0.5446	-0.6533	-1.3097	0.7031	0.6564	-395.5632	-321.9608	0.3951	0.2772
0.5319	0.6179	300	0.5262	-0.6809	-1.4161	0.7344	0.7353	-406.2091	-324.7231	0.4856	0.3738
0.5268	0.8239	400	0.5195	-0.7599	-1.5725	0.7344	0.8126	-421.8436	-332.6288	0.5181	0.3998

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.1+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1