llama-3-instruct-8b-cpo-full-beta-2.5_alpha-0.25_e-version-false-lr-1e-6

This model was trained from scratch on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0998
• Record Ada Margin/mean: -2.2225
• Ref Logratios Sum/mean: 1.0990
• Ref Logratios Sum/std: 39.6734
• Ref Logratios Sum/min: -147.9120
• Ref Logratios Sum/max: 120.8177
• Ref Logratios/mean: 0.0093
• Ref Logratios/std: 0.1420
• Ref Logratios/min: -0.4063
• Ref Logratios/max: 0.6104
• Rewards/chosen: -2.2488
• Rewards/rejected: -3.1221
• Rewards/accuracies: 0.8024
• Rewards/margins: 0.8733
• Logps/rejected: -1.5593
• Logps/chosen: -1.1931
• Logits/rejected: -1.2589
• Logits/chosen: -1.2728

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: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 4
• 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: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Record Ada Margin/mean	Ref Logratios Sum/mean	Ref Logratios Sum/std	Ref Logratios Sum/min	Ref Logratios Sum/max	Ref Logratios/mean	Ref Logratios/std	Ref Logratios/min	Ref Logratios/max	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.0928	0.8547	400	0.0998	-2.2225	1.0990	39.6734	-147.9120	120.8177	0.0093	0.1420	-0.4063	0.6104	-2.2488	-3.1221	0.8024	0.8733	-1.5593	-1.1931	-1.2589	-1.2728

Framework versions

• Transformers 4.44.2
• Pytorch 2.2.2+cu121
• Datasets 2.18.0
• Tokenizers 0.19.0