Llama-3.1-8B-Instruct_dpo_sg_values_p025_OA_silver

This model is a fine-tuned version of meta-llama/Llama-3.1-8B-Instruct on the dpo_sg_values_p025_OA_silver dataset. It achieves the following results on the evaluation set:

• Loss: 0.1344
• Rewards/chosen: -0.3139
• Rewards/rejected: -3.7636
• Rewards/accuracies: 0.9440
• Rewards/margins: 3.4497
• Logps/chosen: -5.2304
• Logps/rejected: -44.2580
• Logits/chosen: -0.7692
• Logits/rejected: -0.7949

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: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/chosen	Logps/rejected	Logits/chosen	Logits/rejected
0.6882	0.1495	250	0.6470	-0.0027	-0.0991	0.8740	0.0963	-2.1189	-7.6131	-0.6586	-0.6706
0.4574	0.2990	500	0.3221	-0.1440	-1.3807	0.8900	1.2367	-3.5317	-20.4297	-0.7456	-0.7637
0.1975	0.4486	750	0.1916	-0.3061	-2.8027	0.9300	2.4966	-5.1523	-34.6491	-0.8112	-0.8331
0.128	0.5981	1000	0.1556	-0.3224	-3.2965	0.9360	2.9741	-5.3155	-39.5871	-0.7987	-0.8227
0.1217	0.7476	1250	0.1421	-0.3065	-3.5741	0.9380	3.2675	-5.1568	-42.3631	-0.7826	-0.8077
0.0864	0.8971	1500	0.1368	-0.3144	-3.7298	0.9440	3.4154	-5.2351	-43.9201	-0.7712	-0.7968

Framework versions

• PEFT 0.15.2
• Transformers 4.49.0
• Pytorch 2.6.0+cu124
• Datasets 2.21.0
• Tokenizers 0.21.1