Mistral-7B-Base-SimPO2-5e-7

This model is a fine-tuned version of alignment-handbook/zephyr-7b-sft-full on the princeton-nlp/mistral-instruct-ultrafeedback dataset. It achieves the following results on the evaluation set:

• Logits/chosen: -2.9000
• Logits/rejected: -2.8928
• Logps/chosen: -1.1346
• Logps/rejected: -1.2466
• Loss: -0.0708
• Rewards/accuracies: 0.5758
• Rewards/chosen: -1.1346
• Rewards/margins: 0.1120
• Rewards/rejected: -1.2466

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: 16
• total_train_batch_size: 128
• total_eval_batch_size: 16
• 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	Logits/chosen	Logits/rejected	Logps/chosen	Logps/rejected	Validation Loss	Rewards/accuracies	Rewards/chosen	Rewards/margins	Rewards/rejected
-0.0905	0.8573	400	-2.9000	-2.8928	-1.1346	-1.2466	-0.0708	0.5758	-1.1346	0.1120	-1.2466

Framework versions

• Transformers 4.42.0
• Pytorch 2.1.2+cu121
• Datasets 2.14.6
• Tokenizers 0.19.1