deepseek-8b-orpo-lora

This model is a fine-tuned version of deepseek-ai/deepseek-math-7b-base on the zfz1/my_preference_gsm8k_deepseek dataset. It achieves the following results on the evaluation set:

• Loss: 0.6818
• Rewards/chosen: -0.0338
• Rewards/rejected: -0.0840
• Rewards/accuracies: 0.8088
• Rewards/margins: 0.0502
• Logps/rejected: -0.8398
• Logps/chosen: -0.3377
• Logits/rejected: 34.4233
• Logits/chosen: 35.5254
• Nll Loss: 0.6414
• Log Odds Ratio: -0.4212
• Log Odds Chosen: 1.0634

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: 0.0001
• train_batch_size: 16
• eval_batch_size: 16
• seed: 43
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 2
• 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: 2

Training results

Framework versions

• PEFT 0.11.1
• Transformers 4.42.3
• Pytorch 2.1.2
• Datasets 2.20.0
• Tokenizers 0.19.1