mistral-7b-bfloat16-dpo-full

This model is a fine-tuned version of ghimiresunil/mistral-7b-bfloat16-trainer-full on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.0000
• Rewards/chosen: 0.1753
• Rewards/rejected: -13.6755
• Rewards/accuracies: 1.0
• Rewards/margins: 13.8507
• Logps/rejected: -283.8056
• Logps/chosen: -297.8054
• Logits/rejected: -2.4650
• Logits/chosen: -2.7869

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: 7
• total_train_batch_size: 28
• total_eval_batch_size: 28
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.0265	1.0	210	0.0000	-0.2153	-13.4824	1.0	13.2671	-281.8752	-301.7108	-2.4639	-2.7959
0.0002	2.0	420	0.0001	0.1765	-12.0072	1.0	12.1837	-267.1233	-297.7928	-2.4774	-2.7947
0.0	3.0	630	0.0000	0.1753	-13.6755	1.0	13.8507	-283.8056	-297.8054	-2.4650	-2.7869

Framework versions

• Transformers 4.35.0
• Pytorch 2.1.1+cu121
• Datasets 2.14.6
• Tokenizers 0.14.1