zephyr-7b-dpo-qlora-fix

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-qlora on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

• Loss: 0.5279
• Rewards/chosen: -1.0268
• Rewards/rejected: -1.8204
• Rewards/accuracies: 0.7617
• Rewards/margins: 0.7936
• Logps/rejected: -429.5990
• Logps/chosen: -349.1275
• Logits/rejected: 1.1048
• Logits/chosen: 1.1977

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: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• total_eval_batch_size: 64
• 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
0.5985	0.21	100	0.6167	-0.6622	-0.9981	0.7031	0.3359	-347.3664	-312.6618	-2.0061	-1.9992
0.5302	0.42	200	0.5495	-0.8758	-1.5987	0.7461	0.7229	-407.4292	-334.0204	0.3116	0.4001
0.533	0.63	300	0.5384	-0.8142	-1.5157	0.7617	0.7016	-399.1313	-327.8605	0.5716	0.6809
0.518	0.84	400	0.5276	-1.0554	-1.8498	0.75	0.7944	-432.5438	-351.9892	1.1053	1.1955

Framework versions

• PEFT 0.7.1
• Transformers 4.36.2
• Pytorch 2.1.2+cu121
• Datasets 2.14.6
• Tokenizers 0.15.1