claude-bot
598f7fd662
Replaces the previous tree entirely. The pipeline is four separable pieces, each
testable on its own:
1. leanbench/llm.py provider-neutral OpenAI-compatible connector, stdlib only,
with retries, JSON coercion and a scripted client so the
rest of the code is testable without a network.
2. leanbench/detect.py per-page detection of formalizable mathematics. Case 1 is
mathematics asserted as fact (definition/theorem/lemma/...,
deliberately not sub-classified); case 2 is a problem left
open to the reader. Hits come back as spans -- start page
and index, end page and index -- so several per page is
normal. A hit cut off by the page end makes the detector
re-read that page together with the next, repeatedly, so an
item spread over three pages is found whole.
Offsets come from verbatim quotes resolved by
leanbench/anchors.py (exact -> whitespace-insensitive ->
fuzzy), because models quote well and count badly.
3. leanbench/formalize.py + leanbench/verify.py
each span becomes one Lean 4 declaration, judged by the real
compiler: full elaboration for case 1, statement legality
only for case 2, which has no proof to check. Lean's own
diagnostics feed back for a repair attempt. "verified" (Lean
accepted it) and "proved" (accepted with no sorry) are kept
strictly apart.
4. leanbench/db.py indexed SQLite plus JSONL export. Every item is indexed by
the types of its inputs, or by its own type when it is not a
proof, and records the source book (sha256) and the page it
starts on -- so `drop --source H --pages A-B` followed by a
re-ingest redoes exactly those pages and nothing else.
Infrastructure:
- lean/ Lake project pinned to Mathlib.
- docker/ image carrying Lean 4 and a prebuilt Mathlib. The Mathlib
revision and the toolchain are resolved against each other at
build time rather than both hardcoded, so the build does not rot.
One Docker layer per Lean step, each long one under `timeout`:
the gigabyte cache fetch earns a layer that a later failure never
invalidates, and a stalled step fails with a message instead of
sitting silent for an hour. There is deliberately no `lake build`
of our own library -- verification runs `lake env lean` against
Mathlib's oleans and needs nothing else, and building the library
was pure cost. docker/smoke.py runs the whole pipeline inside the
image against the real compiler, with the LLM scripted.
- .gitea/workflows ci (lint + tests in a venv, since the runner image's Python is
Debian-managed and refuses installs into itself), docker (build,
verify, optionally push), ingest (submit a book by hand). The
docker build runs with --progress=plain, because BuildKit's
default renderer collapses a long step into one rewritten line
and a stall then looks exactly like progress.
The docker workflow is not a throwaway build: it leaves the image, the layer cache,
an idle `lean-bench` container and the `lean-bench-data` / `lean-bench-books`
volumes in place on the runner, and never prunes, so the environment is there to be
used afterwards. Pushing to the registry is optional and cannot fail the build --
Gitea's container registry does not accept the built-in Actions token, so that needs
a REGISTRY_TOKEN secret with write:package.
58 tests; they need neither network nor Lean.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01Ci5T3gWDhhmqDec3e1XtYZ
80 lines
2.2 KiB
Python
80 lines
2.2 KiB
Python
from __future__ import annotations
|
||
|
||
import json
|
||
from pathlib import Path
|
||
|
||
import pytest
|
||
|
||
from leanbench.document import Document
|
||
from leanbench.verify import MODE_FULL, VerifyResult
|
||
|
||
PAGE_ONE = """Chapter 1. Divisibility
|
||
|
||
1.6 Divisibility in the Integers
|
||
|
||
Definition 1.6.1. Let a and b be integers. We say a divides b, written a | b,
|
||
if there exists an integer k such that b = a k.
|
||
|
||
Theorem 1.6.2. Let a, b and c be integers. If a divides b and a divides c,
|
||
then a divides b + c.
|
||
|
||
Proof. Write b = a k and c = a m. Then b + c = a (k + m). QED
|
||
|
||
Theorem 1.6.3. Let a and b be nonzero integers. Then the set of common
|
||
"""
|
||
|
||
PAGE_TWO = """divisors of a and b has a greatest element, the greatest common divisor.
|
||
|
||
Exercise 1.6.4. Show that if a divides b and b divides a, then a = b or a = -b.
|
||
|
||
Exercise 1.6.5. Prove that the sum of two odd integers is even.
|
||
"""
|
||
|
||
|
||
@pytest.fixture()
|
||
def two_page_book(tmp_path: Path) -> Document:
|
||
path = tmp_path / "book.txt"
|
||
path.write_text(PAGE_ONE + "\f" + PAGE_TWO, encoding="utf-8")
|
||
return Document.load(path, title="Test Book")
|
||
|
||
|
||
class FakeChecker:
|
||
"""A Lean stand-in: fails anything containing ``BAD``, otherwise accepts."""
|
||
|
||
def __init__(self) -> None:
|
||
self.calls: list[str] = []
|
||
|
||
def available(self) -> bool:
|
||
return True
|
||
|
||
def check(self, code: str, *, mode: str = MODE_FULL, imports=None) -> VerifyResult:
|
||
self.calls.append(code)
|
||
if "BAD" in code:
|
||
return VerifyResult(False, mode, False, "LeanBenchCheck.lean:3:0: error: bad")
|
||
return VerifyResult(True, mode, "sorry" in code, "")
|
||
|
||
|
||
@pytest.fixture()
|
||
def fake_checker() -> FakeChecker:
|
||
return FakeChecker()
|
||
|
||
|
||
def detection_reply(items: list[dict]) -> str:
|
||
return json.dumps({"items": items})
|
||
|
||
|
||
def formalize_reply(**overrides) -> str:
|
||
payload = {
|
||
"skip": False,
|
||
"reason": "",
|
||
"imports": ["Mathlib"],
|
||
"name": "test_decl",
|
||
"lean_code": "theorem test_decl (a b : Int) (h : a ∣ b) : a ∣ b := by\n sorry",
|
||
"has_proof": False,
|
||
"informal": "a divides b",
|
||
"index_types": ["Int"],
|
||
"confidence": 0.8,
|
||
}
|
||
payload.update(overrides)
|
||
return json.dumps(payload)
|