Llama-2-7b-hf-DPO-LookAhead-5_TTree1.4_TT0.9_TP0.7_TE0.2_V2

This model is a fine-tuned version of meta-llama/Llama-2-7b-hf on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.2147
• Rewards/chosen: -2.3589
• Rewards/rejected: -2.1848
• Rewards/accuracies: 0.3333
• Rewards/margins: -0.1740
• Logps/rejected: -176.9075
• Logps/chosen: -185.7344
• Logits/rejected: -0.3397
• Logits/chosen: -0.3554

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-05
• train_batch_size: 2
• eval_batch_size: 2
• 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: 10
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.7064	0.3020	77	0.7263	-0.0650	-0.0237	0.5	-0.0414	-155.2957	-162.7962	0.2969	0.2895
0.6816	0.6039	154	0.7127	-0.1015	-0.1222	0.5	0.0207	-156.2813	-163.1606	0.2989	0.2915
0.6192	0.9059	231	0.7010	-0.0808	-0.1624	0.5833	0.0816	-156.6835	-162.9536	0.2774	0.2692
0.2805	1.2078	308	0.8302	-0.5931	-0.6582	0.6667	0.0651	-161.6412	-168.0767	0.1922	0.1839
0.3604	1.5098	385	0.8663	-0.8552	-0.8899	0.5833	0.0347	-163.9578	-170.6977	0.0866	0.0775
0.3524	1.8118	462	0.9587	-1.3495	-1.3440	0.5	-0.0055	-168.4993	-175.6406	-0.0538	-0.0645
0.2168	2.1137	539	1.0785	-1.8309	-1.7601	0.5833	-0.0708	-172.6597	-180.4545	-0.2246	-0.2382
0.0395	2.4157	616	1.2284	-2.4130	-2.2406	0.3333	-0.1724	-177.4654	-186.2757	-0.3472	-0.3633
0.2081	2.7176	693	1.2147	-2.3589	-2.1848	0.3333	-0.1740	-176.9075	-185.7344	-0.3397	-0.3554

Framework versions

• PEFT 0.12.0
• Transformers 4.44.0
• Pytorch 2.4.0+cu121
• Datasets 3.0.2
• Tokenizers 0.19.1