Mistral2_150_STEPS_03beta_1e7rate_CDPOSFT

This model is a fine-tuned version of tsavage68/mistralit2_1000_STEPS_5e7_SFT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5695
• Rewards/chosen: 0.0666
• Rewards/rejected: -0.3152
• Rewards/accuracies: 0.6659
• Rewards/margins: 0.3818
• Logps/rejected: -27.6074
• Logps/chosen: -23.4495
• Logits/rejected: -2.2642
• Logits/chosen: -2.2637

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-07
• train_batch_size: 4
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 100
• training_steps: 150

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6838	0.0977	50	0.6757	0.0819	0.0446	0.6088	0.0374	-26.4083	-23.3987	-2.3067	-2.3063
0.5869	0.1953	100	0.5936	0.1585	-0.1222	0.6418	0.2808	-26.9643	-23.1432	-2.2711	-2.2707
0.5203	0.2930	150	0.5695	0.0666	-0.3152	0.6659	0.3818	-27.6074	-23.4495	-2.2642	-2.2637

Framework versions

• Transformers 4.40.1
• Pytorch 2.0.0+cu117
• Datasets 2.19.1
• Tokenizers 0.19.1