zephyr0-7b-ultra-p-0.02

This model is a fine-tuned version of HuggingFaceH4/mistral-7b-sft-beta on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5172
• Rewards/chosen: -1.0774
• Rewards/rejected: -1.8857
• Rewards/accuracies: 0.75
• Rewards/margins: 0.8083
• Logps/rejected: -416.3554
• Logps/chosen: -330.1051
• Logits/rejected: -1.9304
• 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: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• gradient_accumulation_steps: 8
• 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: linear
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
0.576	0.2060	100	0.5490	-0.7973	-1.3203	0.7266	0.5230	-359.8105	-302.0933	-2.4928	-2.5572
0.5325	0.4119	200	0.5335	-1.0228	-1.6514	0.7266	0.6287	-392.9271	-324.6441	-2.1888	-2.2654
0.5083	0.6179	300	0.5271	-0.9857	-1.7625	0.7422	0.7768	-404.0363	-320.9385	-2.0319	-2.1170
0.5073	0.8239	400	0.5199	-1.0283	-1.8048	0.75	0.7765	-408.2682	-325.1979	-1.9835	-2.0746

Framework versions

• Transformers 4.45.1
• Pytorch 2.4.1+cu121
• Datasets 3.0.0
• Tokenizers 0.20.0