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
91 lines
3.5 KiB
Python
91 lines
3.5 KiB
Python
from __future__ import annotations
|
|
|
|
from leanbench.detect import KIND_PROBLEM, KIND_STATEMENT, Detection
|
|
from leanbench.formalize import formalize, formalize_all, verification_mode
|
|
from leanbench.llm import ScriptedClient
|
|
from leanbench.verify import MODE_FULL, MODE_STATEMENT
|
|
|
|
from .conftest import formalize_reply
|
|
|
|
|
|
def make_detection(kind: str = KIND_STATEMENT) -> Detection:
|
|
return Detection(
|
|
kind=kind,
|
|
label="Theorem 1.6.2",
|
|
summary="divisibility is additive",
|
|
start_page=12,
|
|
start_index=100,
|
|
end_page=12,
|
|
end_index=260,
|
|
text="Theorem 1.6.2. If a divides b and a divides c then a divides b + c.",
|
|
)
|
|
|
|
|
|
def test_open_problems_are_only_checked_as_statements() -> None:
|
|
assert verification_mode(make_detection(KIND_PROBLEM)) == MODE_STATEMENT
|
|
assert verification_mode(make_detection(KIND_STATEMENT)) == MODE_FULL
|
|
|
|
|
|
def test_formalize_produces_verified_code_and_types(fake_checker) -> None:
|
|
client = ScriptedClient([formalize_reply()])
|
|
result = formalize(client, make_detection(), checker=fake_checker)
|
|
assert result.lean_code.startswith("theorem test_decl")
|
|
assert result.verified
|
|
assert not result.proved, "a sorry in the code means it is not proved"
|
|
assert ("Int", "input") in result.index_types
|
|
|
|
|
|
def test_a_real_proof_counts_as_proved(fake_checker) -> None:
|
|
client = ScriptedClient(
|
|
[formalize_reply(lean_code="theorem t (n : Nat) : n = n := rfl", has_proof=True)]
|
|
)
|
|
result = formalize(client, make_detection(), checker=fake_checker)
|
|
assert result.has_proof and result.proved
|
|
|
|
|
|
def test_lean_diagnostics_drive_a_repair_attempt(fake_checker) -> None:
|
|
client = ScriptedClient(
|
|
[
|
|
formalize_reply(lean_code="theorem t : BAD := by sorry"),
|
|
formalize_reply(lean_code="theorem t (n : Nat) : n = n := rfl", has_proof=True),
|
|
]
|
|
)
|
|
result = formalize(client, make_detection(), checker=fake_checker, repair_attempts=1)
|
|
assert result.attempts == 2
|
|
assert result.verified
|
|
assert len(client.calls) == 2
|
|
repair_prompt = client.calls[1][-1]["content"]
|
|
assert "error: bad" in repair_prompt, "the compiler's own words go back to the model"
|
|
|
|
|
|
def test_repair_gives_up_after_the_budget(fake_checker) -> None:
|
|
client = ScriptedClient([formalize_reply(lean_code="theorem t : BAD := by sorry")] * 3)
|
|
result = formalize(client, make_detection(), checker=fake_checker, repair_attempts=1)
|
|
assert result.attempts == 2
|
|
assert not result.verified
|
|
|
|
|
|
def test_model_may_decline_an_unformalizable_passage(fake_checker) -> None:
|
|
client = ScriptedClient(['{"skip": true, "reason": "prose only"}'])
|
|
result = formalize(client, make_detection(), checker=fake_checker)
|
|
assert result.skipped and result.skip_reason == "prose only"
|
|
assert fake_checker.calls == []
|
|
|
|
|
|
def test_imports_inside_the_body_are_stripped(fake_checker) -> None:
|
|
client = ScriptedClient(
|
|
[formalize_reply(lean_code="import Mathlib\ntheorem t (n : Nat) : n = n := rfl")]
|
|
)
|
|
result = formalize(client, make_detection(), checker=fake_checker)
|
|
assert "import" not in result.lean_code
|
|
|
|
|
|
def test_one_bad_item_does_not_stop_the_batch(fake_checker) -> None:
|
|
client = ScriptedClient(["garbage, not json", formalize_reply()])
|
|
results = formalize_all(
|
|
client, [make_detection(), make_detection(KIND_PROBLEM)], checker=fake_checker
|
|
)
|
|
assert len(results) == 2
|
|
assert results[0].skipped and "formalizer failed" in results[0].skip_reason
|
|
assert results[1].verified
|