nash_dpo_merge_iter_3

This model is a fine-tuned version of YYYYYYibo/nash_dpo_merge_iter_2 on the updated and the original datasets. It achieves the following results on the evaluation set:

• Loss: 0.5506
• Rewards/chosen: -0.4423
• Rewards/rejected: -0.9695
• Rewards/accuracies: 0.7060
• Rewards/margins: 0.5272
• Logps/rejected: -386.9142
• Logps/chosen: -353.7201
• Logits/rejected: 0.3453
• Logits/chosen: -0.2965

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-06
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 16
• total_train_batch_size: 128
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.5641	0.49	100	0.5612	-0.5147	-0.9965	0.7020	0.4819	-389.6182	-360.9578	0.4191	-0.1955
0.5515	0.98	200	0.5506	-0.4423	-0.9695	0.7060	0.5272	-386.9142	-353.7201	0.3453	-0.2965

Framework versions

• PEFT 0.7.1
• Transformers 4.36.2
• Pytorch 2.1.2+cu121
• Datasets 2.14.6
• Tokenizers 0.15.2