zephyr-dpo-qlora-uf-ours-5e-7-epoch1

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the generation/UF dataset. It achieves the following results on the evaluation set:

• Loss: 0.6853
• Rewards/chosen: 0.0412
• Rewards/rejected: 0.0193
• Rewards/accuracies: 0.5880
• Rewards/margins: 0.0219
• Rewards/margins Max: 0.1290
• Rewards/margins Min: -0.0714
• Rewards/margins Std: 0.0669
• Logps/rejected: -256.6506
• Logps/chosen: -280.4748
• Logits/rejected: -2.7413
• Logits/chosen: -2.7795

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-07
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• total_eval_batch_size: 16
• 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	Rewards/margins Max	Rewards/margins Min	Rewards/margins Std	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.6803	0.28	100	0.6908	0.0154	0.0089	0.5800	0.0065	0.0406	-0.0225	0.0210	-257.6936	-283.0562	-2.7654	-2.8039
0.6568	0.56	200	0.6869	0.0391	0.0218	0.5960	0.0173	0.1025	-0.0572	0.0531	-256.3947	-280.6796	-2.7494	-2.7877
0.6398	0.85	300	0.6854	0.0408	0.0192	0.5900	0.0216	0.1281	-0.0713	0.0665	-256.6587	-280.5105	-2.7434	-2.7814

Framework versions

• PEFT 0.7.1
• Transformers 4.39.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.14.6
• Tokenizers 0.15.2