Na_L3_150steps_1e6rate_01beta_cSFTDPO

This model is a fine-tuned version of tsavage68/Na_L3_100steps_1e6rate_SFT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0000
• Rewards/chosen: 1.3563
• Rewards/rejected: -9.7135
• Rewards/accuracies: 1.0
• Rewards/margins: 11.0698
• Logps/rejected: -138.6369
• Logps/chosen: -11.3269
• Logits/rejected: -0.9253
• Logits/chosen: -0.8531

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: 2
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 4
• 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
0.0004	0.2667	50	0.0002	1.1139	-7.3111	1.0	8.4250	-114.6128	-13.7514	-0.9426	-0.8757
0.0	0.5333	100	0.0000	1.3245	-9.2705	1.0	10.5951	-134.2070	-11.6448	-0.9282	-0.8570
0.0	0.8	150	0.0000	1.3563	-9.7135	1.0	11.0698	-138.6369	-11.3269	-0.9253	-0.8531

Framework versions

• Transformers 4.44.2
• Pytorch 2.4.0+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1