zephyr-7b-dpo-lora

This model is a fine-tuned version of HuggingFaceH4/mistral-7b-sft-beta on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5082
• Rewards/chosen: 0.0025
• Rewards/rejected: -0.9047
• Rewards/accuracies: 0.7222
• Rewards/margins: 0.9072
• Logps/rejected: -276.6852
• Logps/chosen: -271.8461
• Logits/rejected: -2.7167
• Logits/chosen: -2.7365

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: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• total_eval_batch_size: 32
• 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
0.5795	0.1047	100	0.5875	0.0265	-0.3721	0.6825	0.3986	-271.3593	-271.6063	-2.7688	-2.7900
0.5449	0.2094	200	0.5520	0.0601	-0.5726	0.7103	0.6327	-273.3645	-271.2704	-2.7792	-2.7981
0.545	0.3141	300	0.5320	-0.0197	-0.7637	0.7044	0.7439	-275.2751	-272.0686	-2.7616	-2.7803
0.4747	0.4187	400	0.5228	-0.1728	-0.9527	0.7004	0.7798	-277.1651	-273.5996	-2.7532	-2.7732
0.5367	0.5234	500	0.5175	-0.2142	-1.0435	0.7143	0.8293	-278.0737	-274.0135	-2.7339	-2.7540
0.5031	0.6281	600	0.5139	-0.2939	-1.1329	0.7024	0.8389	-278.9670	-274.8105	-2.7071	-2.7268
0.5057	0.7328	700	0.5084	-0.0108	-0.9049	0.7202	0.8941	-276.6876	-271.9794	-2.7207	-2.7404
0.5172	0.8375	800	0.5090	-0.0300	-0.9231	0.7183	0.8931	-276.8697	-272.1711	-2.7173	-2.7371
0.5173	0.9422	900	0.5084	-0.0008	-0.9080	0.7222	0.9072	-276.7181	-271.8789	-2.7174	-2.7372

Framework versions

• PEFT 0.7.1
• Transformers 4.45.2
• Pytorch 2.4.0+cu121
• Datasets 2.14.6
• Tokenizers 0.20.1