zephyr-7b-ipo-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: 18.3397
• Rewards/chosen: 0.0292
• Rewards/rejected: -0.1006
• Rewards/accuracies: 0.7200
• Rewards/margins: 0.1298
• Logps/rejected: -212.0379
• Logps/chosen: -255.2319
• Logits/rejected: -1.7967
• Logits/chosen: -2.0243

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
• num_devices: 4
• gradient_accumulation_steps: 32
• total_train_batch_size: 256
• total_eval_batch_size: 16
• 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
19.3937	1.0	242	19.3450	0.0291	-0.0729	0.7040	0.1020	-211.7608	-255.2333	-1.7962	-2.0237
19.376	2.0	484	18.8198	0.0270	-0.0949	0.7020	0.1218	-211.9809	-255.2546	-1.7954	-2.0232
18.4503	3.0	726	18.3397	0.0292	-0.1006	0.7200	0.1298	-212.0379	-255.2319	-1.7967	-2.0243

Framework versions

• Transformers 4.35.0
• Pytorch 2.1.2+cu121
• Datasets 2.14.6
• Tokenizers 0.14.1