selective-pairrm-33045197-mt0

This model is a fine-tuned version of mistralai/Mistral-7B-Instruct-v0.2 on the snorkelai/Snorkel-Mistral-PairRM-DPO-Dataset dataset. It achieves the following results on the evaluation set:

• Loss: 0.6825
• Rewards/chosen: -0.2329
• Rewards/rejected: -0.2692
• Rewards/accuracies: 0.6055
• Rewards/margins: 0.0362
• Logps/rejected: -417.6746
• Logps/chosen: -401.4102
• Logits/rejected: -3.1643
• Logits/chosen: -3.1708

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: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• total_eval_batch_size: 32
• 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
0.6786	0.32	100	0.6868	-0.0869	-0.1015	0.5547	0.0146	-400.9028	-386.8027	-2.8786	-2.8855
0.6615	0.64	200	0.6828	-0.1851	-0.2144	0.5938	0.0294	-412.2021	-396.6207	-3.0607	-3.0672
0.6539	0.96	300	0.6821	-0.2322	-0.2693	0.6055	0.0371	-417.6892	-401.3395	-3.1645	-3.1709

Framework versions

• Transformers 4.36.2
• Pytorch 2.1.2
• Datasets 2.14.6
• Tokenizers 0.15.0