Na_L3_350steps_1e7rate_05beta_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: 0.0001
• Rewards/chosen: 1.2270
• Rewards/rejected: -8.7510
• Rewards/accuracies: 1.0
• Rewards/margins: 9.9780
• Logps/rejected: -59.0037
• Logps/chosen: -22.4361
• Logits/rejected: -0.9548
• Logits/chosen: -0.8881

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: 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: 350

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.501	0.2667	50	0.4173	0.1064	-0.5742	1.0	0.6806	-42.6500	-24.6773	-0.9522	-0.8855
0.0047	0.5333	100	0.0017	0.6482	-5.8984	1.0	6.5466	-53.2984	-23.5937	-0.9531	-0.8872
0.0001	0.8	150	0.0001	1.1509	-8.2919	1.0	9.4428	-58.0856	-22.5883	-0.9550	-0.8887
0.0001	1.0667	200	0.0001	1.2083	-8.6310	1.0	9.8393	-58.7638	-22.4736	-0.9544	-0.8879
0.0001	1.3333	250	0.0001	1.2270	-8.7563	1.0	9.9833	-59.0143	-22.4362	-0.9536	-0.8869
0.0001	1.6	300	0.0001	1.2204	-8.7546	1.0	9.9751	-59.0110	-22.4493	-0.9548	-0.8881
0.0001	1.8667	350	0.0001	1.2270	-8.7510	1.0	9.9780	-59.0037	-22.4361	-0.9548	-0.8881

Framework versions

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