zephyr-7b-ipo-lora-5ep

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: 17.9243
• Rewards/chosen: 0.0274
• Rewards/rejected: -0.1133
• Rewards/accuracies: 0.7360
• Rewards/margins: 0.1407
• Logps/rejected: -212.1647
• Logps/chosen: -255.2498
• Logits/rejected: -1.7956
• Logits/chosen: -2.0232

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: 5

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
19.7198	1.0	242	19.4561	0.0339	-0.0669	0.6980	0.1008	-211.7013	-255.1847	-1.7967	-2.0240
19.4456	2.0	484	18.4657	0.0251	-0.0989	0.7140	0.1240	-212.0212	-255.2727	-1.7950	-2.0227
18.2488	3.0	726	18.0217	0.0246	-0.1117	0.7400	0.1362	-212.1486	-255.2784	-1.7960	-2.0237
17.9448	4.0	968	17.9649	0.0243	-0.1137	0.7380	0.1379	-212.1687	-255.2813	-1.7954	-2.0230
17.8013	5.0	1210	17.9243	0.0274	-0.1133	0.7360	0.1407	-212.1647	-255.2498	-1.7956	-2.0232

Framework versions

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