miniCTX: Neural Theorem Proving with (Long-)Contexts
File-tuned context model from miniCTX: Neural Theorem Proving with (Long-)Contexts.
- Base language model:
deepseek-ai/deepseek-coder-1.3b-base - Data: ntp-mathlib-instruct-context
It is specifically finetuned for Lean 4 tactic prediction given proof states and optional file contexts.
Performance
Please see our paper.
Example input
/- You are proving a theorem in Lean 4.
You are given the following information:
- The file contents up to the current tactic, inside [CTX]...[/CTX]
- The current proof state, inside [STATE]...[/STATE]
Your task is to generate the next tactic in the proof.
Put the next tactic inside [TAC]...[/TAC]
-/
[CTX]
import Mathlib.Data.Nat.Prime
theorem test_thm (m n : Nat) (h : m.Coprime n) : m.gcd n = 1 := by
[/CTX]
[STATE]
m n : ℕ
h : Nat.Coprime m n
⊢ Nat.gcd m n = 1
[/STATE]
[TAC]
Example output
rw [Nat.Coprime] at h
[/TAC]
Citation
Please cite:
@misc{hu2024minictx,
title={miniCTX: Neural Theorem Proving with (Long-)Contexts},
author={Jiewen Hu and Thomas Zhu and Sean Welleck},
year={2024},
eprint={2408.03350},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2408.03350},
}