LLM Tuning Patterns

关于

LLM Tuning Patterns provides a research-driven framework for optimizing Large Language Model performance across diverse domains like formal theorem proving, code generation, and creative exploration. Drawing from APOLLO and Godel-Prover research, this skill implements specific guidance for parameters such as temperature, top_p, and token limits to ensure maximum reasoning capability. It helps users avoid common pitfalls like truncated chain-of-thought and overly deterministic proof paths, while introducing advanced techniques like parallel sampling and proof-plan-first prompting.

主要功能

• 0 GitHub stars
• Parallel sampling and best-of-N selection strategies
• Optimized settings for Lean 4 formal theorem proving
• Chain-of-thought proof planning templates
• Research-backed guidance for temperature and top_p scaling
• Evidence-based parameter configurations for specific LLM tasks

使用场景

• Improving success rates in formal verification and Lean 4 tactic generation
• Scaling reasoning depth for complex architectural or creative exploration tasks
• Configuring deterministic and efficient parameters for code generation