doplhin-mistral-dpo-ultrafeedback-binarized-preferences-hinge

This model is a fine-tuned version of cognitivecomputations/dolphin-2.1-mistral-7b on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5904
• Rewards/chosen: -3.5980
• Rewards/rejected: -6.3560
• Rewards/accuracies: 0.7954
• Rewards/margins: 2.7581
• Logps/rejected: -378.3953
• Logps/chosen: -389.8724
• Logits/rejected: -2.4206
• Logits/chosen: -2.4999

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-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• 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	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.9205	0.25	700	0.7331	-4.0569	-6.3048	0.7489	2.2478	-377.8824	-394.4619	-2.5742	-2.6507
0.6493	0.51	1400	0.6532	-4.0240	-7.1929	0.7889	3.1689	-386.7637	-394.1322	-2.4562	-2.5463
0.663	0.76	2100	0.5904	-3.5980	-6.3560	0.7954	2.7581	-378.3953	-389.8724	-2.4206	-2.4999

Framework versions

• PEFT 0.8.2
• Transformers 4.37.2
• Pytorch 2.1.2+cu121
• Datasets 2.16.1
• Tokenizers 0.15.2