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: 18.0888
• Rewards/chosen: 0.0201
• Rewards/rejected: -0.1194
• Rewards/accuracies: 0.7320
• Rewards/margins: 0.1395
• Logps/rejected: -212.8609
• Logps/chosen: -268.5415
• Logits/rejected: -2.3003
• Logits/chosen: -2.6873

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: 1
• eval_batch_size: 1
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 32
• total_train_batch_size: 256
• total_eval_batch_size: 8
• 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
19.1303	1.0	242	18.9124	0.0302	-0.0939	0.7360	0.1241	-212.6061	-268.4410	-2.3029	-2.6884
19.0824	2.0	484	18.2266	0.0186	-0.1163	0.7600	0.1349	-212.8297	-268.5567	-2.3011	-2.6877
18.0366	3.0	726	18.0888	0.0201	-0.1194	0.7320	0.1395	-212.8609	-268.5415	-2.3003	-2.6873

Framework versions

• Transformers 4.36.1
• Pytorch 2.1.2+cu121
• Datasets 2.14.6
• Tokenizers 0.15.0

Training procedure

Framework versions

• PEFT 0.6.1