zephyr-7b-mypo3-qlora-beta2.5-lr5e6

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.3495
• Rewards/chosen: 0.0059
• Rewards/rejected: -0.3228
• Rewards/accuracies: 0.7422
• Rewards/margins: 0.3287
• Logps/rejected: -1.1745
• Logps/chosen: -0.8887
• Logits/rejected: -1.9998
• Logits/chosen: -2.0873

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: 8
• eval_batch_size: 16
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 128
• total_eval_batch_size: 64
• 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
1.3555	0.2093	100	1.3642	-0.0272	-0.2716	0.7266	0.2444	-1.1541	-0.9020	-2.0549	-2.1403
1.355	0.4186	200	1.3551	0.0565	-0.2069	0.7227	0.2634	-1.1282	-0.8685	-2.0756	-2.1581
1.344	0.6279	300	1.3538	-0.0407	-0.3700	0.7363	0.3293	-1.1934	-0.9073	-2.0130	-2.0998
1.3506	0.8373	400	1.3495	0.0059	-0.3228	0.7422	0.3287	-1.1745	-0.8887	-1.9998	-2.0873

Framework versions

• PEFT 0.10.0
• Transformers 4.43.1
• Pytorch 2.1.2+cu121
• Datasets 2.18.0
• Tokenizers 0.19.1