Model Overview

CorticalStack/pastiche-crown-clown-7b-dare-dpo is a 7 billion parameter language model developed by CorticalStack. This model is a Direct Preference Optimization (DPO) fine-tuned variant of the base pastiche-crown-clown-7b-dare model. The fine-tuning process leveraged the jondurbin/truthy-dpo-v0.1 dataset to enhance its performance and alignment.

Key Training Details

• Base Model: CorticalStack/pastiche-crown-clown-7b-dare
• Fine-tuning Method: Direct Preference Optimization (DPO)
• Dataset: jondurbin/truthy-dpo-v0.1
• LoRA Configuration:
  • r: 16
  • LoRA alpha: 16
  • LoRA dropout: 0.05
• Training Arguments:
  • Batch size: 4
  • Gradient accumulation steps: 4
  • Optimizer: paged_adamw_32bit
  • Max steps: 200
  • Learning rate: 5e-05
  • Learning rate scheduler type: cosine
  • Beta: 0.1
  • Max prompt length: 1024
  • Max length: 1536

Intended Use

This model is designed for applications where improved response quality and alignment, achieved through DPO fine-tuning, are beneficial. Its training on a 'truthy' dataset suggests potential for tasks requiring factual accuracy and coherent generation.