Na_M2_350steps_1e8rate_03beta_cSFTDPO

This model is a fine-tuned version of tsavage68/Na_M2_1000steps_1e7_SFT on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5980
• Rewards/chosen: 0.0777
• Rewards/rejected: -0.1272
• Rewards/accuracies: 0.9700
• Rewards/margins: 0.2049
• Logps/rejected: -80.3475
• Logps/chosen: -47.8734
• Logits/rejected: -2.5363
• Logits/chosen: -2.5489

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-08
• 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
0.6955	0.2667	50	0.6882	0.0099	-0.0031	0.5600	0.0130	-79.9338	-48.0995	-2.5354	-2.5481
0.6761	0.5333	100	0.6730	0.0130	-0.0315	0.6600	0.0445	-80.0283	-48.0889	-2.5363	-2.5489
0.625	0.8	150	0.6053	0.0511	-0.1371	0.9600	0.1882	-80.3805	-47.9621	-2.5355	-2.5481
0.621	1.0667	200	0.5982	0.0431	-0.1619	0.9700	0.2049	-80.4629	-47.9889	-2.5351	-2.5477
0.593	1.3333	250	0.5948	0.0800	-0.1327	0.9700	0.2127	-80.3656	-47.8657	-2.5362	-2.5489
0.5754	1.6	300	0.5980	0.0777	-0.1272	0.9700	0.2049	-80.3475	-47.8734	-2.5363	-2.5489
0.5882	1.8667	350	0.5980	0.0777	-0.1272	0.9700	0.2049	-80.3475	-47.8734	-2.5363	-2.5489

Framework versions

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