Choose LLM settings for theorem proving, code generation, and creative exploration. It provides examples for output limits and response variety, plus proof planning prompts and a method for generating multiple proof attempts and selecting the best candidate.
한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.
LLM Tuning Patterns
설치하기 전에 에이전트에게 전달되는 지시 내용을 확인할 수 있습니다.
Evidence-based patterns for configuring LLM parameters, based on APOLLO and Godel-Prover research.
Different tasks require different LLM configurations. Use these evidence-based settings.
Based on APOLLO parity analysis:
| Parameter | Value | Rationale |
|---|---|---|
| max_tokens | 4096 | Proofs need space for chain-of-thought |
| temperature | 0.6 | Higher creativity for tactic exploration |
| top_p | 0.95 | Allow diverse proof paths |
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.
For hard proofs, use parallel sampling:
| Parameter | Value | Rationale |
|---|---|---|
| max_tokens | 2048 | Sufficient for most functions |
| temperature | 0.2-0.4 | Prefer deterministic output |
| Parameter | Value | Rationale |
|---|---|---|
| max_tokens | 4096 | Space for exploration |
| temperature | 0.8-1.0 | Maximum creativity |
아직 리뷰가 없습니다. 사용해 본 소감을 남겨 주세요.
로그인 없이 쓸 수 있습니다. 사용 후기, 질문, 작동하지 않은 환경 등을 편하게 남겨 주세요.
아직 댓글이 없습니다. 첫 댓글을 남겨 보세요.
Coordinate background agents through saved files, keeping lengthy work transcripts out of the main conversation. It defines how to pass results between agents, monitor progress, and verify completed work with tests.
한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.
Delegate implementation, file reading, and testing to separate agents, then bring concise summaries back to the main conversation. The rules explain when delegation helps and when a small edit is better handled directly.
한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.
Organize implementation work across research, planning, validation, implementation, and review agents. Each stage passes files to the next, while implementation starts with failing tests and the main conversation coordinates the workflow.
한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.
Connect Agentica SDK agents to Claude Code CLI through a proxy, an intermediary server. Use setup and debugging guidance to address file permissions, streamed responses, result formatting, and agents reporting success without performing the requested operations.
한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.
Consult Agentica APIs when building workflows where multiple AI agents cooperate. This reference covers eleven coordination patterns, agent tracking, shared state, handoffs, and ways to combine results, with pointers to detailed API specifications.
한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.
Write clearer instructions for Agentica and REPL agents by specifying roles, file operations, and output formats. Includes templates for dividing research, coordinating specialists, reviewing solutions, and passing results between agents through files.
한국어 요약은 준비 중입니다. 영어 요약을 표시하고 있습니다.