mcpField
L
llm-tuning-patterns

llm-tuning-patterns

github.com/parcadeiAgent workflows3.9k480 installsUpdated 39w ago
Repository

About llm-tuning-patterns

llm-tuning-patterns is a skill for choosing LLM settings by task, drawing on APOLLO and Godel-Prover research. It provides parameter guidance and prompting patterns for theorem proving, code generation, and creative exploration.

Key features

  1. 01Task-specific LLM parameter guidance
  2. 02Theorem-proving settings and proof-plan prompt
  3. 03Parallel sampling for difficult proofs
  4. 04Code-generation settings
  5. 05Creative-task settings and anti-patterns

Use cases

  1. 01Configuring LLMs for theorem proving
  2. 02Generating Lean 4 tactics from a proof plan
  3. 03Tuning settings for code generation
  4. 04Choosing parameters for creative exploration

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/parcadei/continuous-claude-v3 --skill llm-tuning-patterns

After installing, type/llm-tuning-patternsin Claude Code to confirm it loaded.

Details

From the project README

LLM Tuning Patterns

Evidence-based patterns for configuring LLM parameters, based on APOLLO and Godel-Prover research.

Pattern

Different tasks require different LLM configurations. Use these evidence-based settings.

Theorem Proving / Formal Reasoning

Based on APOLLO parity analysis:

ParameterValueRationale
max_tokens4096Proofs need space for chain-of-thought
temperature0.6Higher creativity for tactic exploration
top_p0.95Allow diverse proof paths
Proof Plan Prompt

Always request a proof plan before tactics:

Given the theorem to prove:
[theorem statement]

First, write a high-level proof plan explaining your approach.
Then, suggest Lean 4 tactics to implement each step.

The proof plan (chain-of-thought) significantly improves tactic quality.

Parallel Sampling

For hard proofs, use parallel sampling:

  • Generate N=8-32 candidate proof attempts
  • Use best-of-N selection
  • Each sample at temperature 0.6-0.8

Code Generation

ParameterValueRationale
max_tokens2048Sufficient for most functions
temperature0.2-0.4Prefer deterministic output

Creative / Exploration Tasks

ParameterValueRationale
max_tokens4096Space for exploration
temperature0.8-1.0Maximum creativity

Anti-Patterns

  • Too low tokens for proofs: 512 tokens truncates chain-of-thought
  • Too low temperature for proofs: 0.2 misses creative tactic paths
  • No proof plan: Jumping to tactics without planning reduces success rate

Source Sessions

  • This session: APOLLO parity - increased max_tokens 512->4096, temp 0.2->0.6
  • This session: Added proof plan prompt for chain-of-thought before tactics

Comments

Markdown supported