zephyr-7b-dpo-lora

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5261
• Rewards/chosen: -0.1563
• Rewards/rejected: -0.9035
• Rewards/accuracies: 0.7315
• Rewards/margins: 0.7472
• Logps/rejected: -232.6935
• Logps/chosen: -266.1220
• Logits/rejected: -1.9669
• Logits/chosen: -2.0211

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: 2
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 32
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 3

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.5647	1.0	968	0.5544	-0.1110	-0.6475	0.7130	0.5365	-230.1329	-265.6688	-2.0148	-2.0684
0.5569	2.0	1936	0.5303	-0.1538	-0.8646	0.7320	0.7108	-232.3041	-266.0967	-1.9784	-2.0322
0.5188	3.0	2904	0.5261	-0.1563	-0.9035	0.7315	0.7472	-232.6935	-266.1220	-1.9669	-2.0211

Framework versions

• Transformers 4.35.0
• Pytorch 2.1.1+cu121
• Datasets 2.14.6
• Tokenizers 0.14.1