pixcell / scripts /pack_core_v1_source.py
aadarwal's picture
Publish PixCell core-v1
7227bd7 verified
Raw
History Blame
18.5 kB
#!/usr/bin/env python3
"""Create the deterministic, minimal PixCell core-v1 source bundle.
This maintainer-only command is the one-time bridge from the original audited
source banks to the standalone public factory. The produced bundle contains
only files consumed by ``build_release.py``:
* accepted-only manifests and audit reports;
* executable per-card generator programs;
* exact model and target images;
* explicit factor/parameter specifications; and
* calibration, topology, and admission evidence used by the release gate.
Historical proposals, rejected candidates, GDS intermediates, caches, and
private benchmark material are intentionally excluded.
"""
from __future__ import annotations
import argparse
import gzip
import hashlib
import io
import json
import tarfile
from pathlib import Path
from typing import Any
from _shared import LEVELS, ReleaseError, read_json, sha256_file, write_json
CORE_NAME = "core-v1"
EXPECTED_RELEASE_DIGEST = (
"e1c509dca337ce485efcb2b35101a052e06b12331b63a2f21a36c93b3682ea89"
)
SOURCE_COMMIT = "2ab0e551fc75c2ad7087ef88ecad2b5de279704b"
BANK_NAMES = {
"l0": "pool_v8_l0_bank",
"l1": "pool_v8_l1_bank",
"l2": "pool_v8_l2_bank",
"l3": "pool_v8_l3_full120",
"l4": "pool_v8_l4_core108",
}
EXPECTED_LEVELS = {
"l0": {
"rows": 177,
"source_digest": (
"495b2c8f4864199213a888a5f8323db08525c76213552f919e901e386784ea76"
),
},
"l1": {
"rows": 215,
"source_digest": (
"30765ca9366965807b65e321a7492afe6d2f8e7ae0c9fbc3a629e4641a771bc1"
),
},
"l2": {
"rows": 118,
"source_digest": (
"05737221cfc0aeb3ef08efc64504d3db031530f5cef1496a48b0239ad23d6b46"
),
},
"l3": {
"rows": 120,
"source_digest": (
"aa696edde9a6bc5ee0c5275d9a7f05eed2a1bc56c600791ebb5c51a3d4953fc6"
),
},
"l4": {
"rows": 108,
"source_digest": (
"8db07b0730b3864a3f864abf08eb5fcf94c144ffcea5d3224590560d429c836f"
),
},
}
OPTIONAL_SIDECARS = (
"parameters.json",
"topology_contract.json",
"factor_assignment.json",
"candidate_evidence.json",
)
ARCHIVE_PAYLOAD = Path | bytes
PRIVATE_MANIFEST_KEYS = frozenset(
{
"proposal_workspace",
"source_root",
"repository_root",
}
)
def _sha256_bytes(payload: bytes) -> str:
return hashlib.sha256(payload).hexdigest()
def _payload_bytes(value: ARCHIVE_PAYLOAD) -> bytes:
return value if isinstance(value, bytes) else value.read_bytes()
def _payload_size(value: ARCHIVE_PAYLOAD) -> int:
return len(value) if isinstance(value, bytes) else value.stat().st_size
def _sanitized_manifest_bytes(value: Any) -> bytes:
"""Remove unused private workspace pointers from a public source capsule."""
def clean(item: Any) -> Any:
if isinstance(item, dict):
return {
str(key): clean(child)
for key, child in item.items()
if str(key) not in PRIVATE_MANIFEST_KEYS
}
if isinstance(item, list):
return [clean(child) for child in item]
return item
return (
json.dumps(
clean(value),
indent=2,
sort_keys=True,
ensure_ascii=True,
allow_nan=False,
)
+ "\n"
).encode("utf-8")
def _inside(root: Path, relative: str, *, label: str) -> Path:
candidate = (root / relative).resolve()
try:
candidate.relative_to(root.resolve())
except ValueError as exc:
raise ReleaseError(f"{label} escapes source bank: {relative}") from exc
if not candidate.is_file():
raise ReleaseError(f"{label} does not exist: {relative}")
return candidate
def _record_files(
bank: Path,
record: dict[str, Any],
) -> dict[str, Path]:
card_relative = record.get("path")
if not isinstance(card_relative, str):
raise ReleaseError(f"{record.get('candidate_id')}: missing card path")
card_dir = (bank / card_relative).resolve()
try:
card_dir.relative_to(bank.resolve())
except ValueError as exc:
raise ReleaseError(f"card path escapes bank: {card_relative}") from exc
if not card_dir.is_dir():
raise ReleaseError(f"missing card directory: {card_relative}")
files = {
"program": _inside(
bank,
str(record.get("program_path") or f"{card_relative}/ground_truth.py"),
label="program",
),
"model_view": _inside(
bank,
str(record.get("model_view_path") or f"{card_relative}/model_view.png"),
label="model view",
),
"target": _inside(
bank,
f"{card_relative}/device_bw.png",
label="target",
),
"calibration": _inside(
bank,
f"{card_relative}/calibration.json",
label="calibration",
),
}
for filename in OPTIONAL_SIDECARS:
path = card_dir / filename
if path.is_file():
files[filename.removesuffix(".json")] = path
if not isinstance(record.get("footprint_um"), list):
path = card_dir / "footprint.json"
if path.is_file():
files["footprint"] = path
return files
def _public_role(record: dict[str, Any]) -> str:
role = str(record.get("variation_role") or record.get("public_role") or "semantic")
return "semantic" if role == "semantic_base" else role
def _first(record: dict[str, Any], *keys: str) -> str:
for key in keys:
value = record.get(key)
if value:
return str(value)
return ""
def _family(level: str, record: dict[str, Any]) -> str:
return (
_first(record, "family", "composition_family")
or {
"l0": "Primitive catalog",
"l1": "Primitive operations",
"l2": "Composed operations",
"l3": "Hierarchical representations",
"l4": "Photonic components",
}[level]
)
def collect_source(
source_root: Path,
) -> tuple[dict[str, ARCHIVE_PAYLOAD], dict[str, Any], dict[str, Any]]:
"""Return archive payload paths, source inventory metadata, and grammar map."""
tasks_root = source_root / "rl" / "tasks"
if not tasks_root.is_dir():
raise ReleaseError(f"source root has no rl/tasks directory: {source_root}")
payload: dict[str, ARCHIVE_PAYLOAD] = {}
grammar_rows: list[dict[str, Any]] = []
level_summaries: dict[str, Any] = {}
curriculum_order = 0
for level in LEVELS:
bank_name = BANK_NAMES[level]
bank = tasks_root / bank_name
manifest_path = bank / "manifest.json"
audit_path = bank / "audit_report.json"
if not manifest_path.is_file() or not audit_path.is_file():
raise ReleaseError(f"{level}: incomplete source bank at {bank}")
manifest = read_json(manifest_path)
audit = read_json(audit_path)
records = manifest.get("records")
if not isinstance(records, list):
raise ReleaseError(f"{level}: manifest has no records")
expected = EXPECTED_LEVELS[level]
if len(records) != expected["rows"]:
raise ReleaseError(
f"{level}: row count {len(records)} != {expected['rows']}"
)
source_digest = manifest.get("dataset_digest_sha256")
if source_digest != expected["source_digest"]:
raise ReleaseError(f"{level}: source digest drift")
if audit.get("dataset_digest_sha256") != source_digest:
raise ReleaseError(f"{level}: manifest/audit digest mismatch")
if "accepted_only" not in str(manifest.get("status", "")):
raise ReleaseError(f"{level}: source bank is not accepted-only")
prefix = Path("rl") / "tasks" / bank_name
payload[(prefix / "manifest.json").as_posix()] = _sanitized_manifest_bytes(
manifest
)
payload[(prefix / "audit_report.json").as_posix()] = audit_path
role_counts: dict[str, int] = {}
for record in records:
files = _record_files(bank, record)
for path in files.values():
relative = path.relative_to(bank)
payload[(prefix / relative).as_posix()] = path
card_id = _first(record, "card_id", "candidate_id", "task_id")
grammar_id = str(record.get("grammar_id") or card_id)
role = _public_role(record)
role_counts[role] = role_counts.get(role, 0) + 1
factor_path = files.get("factor_assignment")
parameter_path = files.get("parameters")
grammar_rows.append(
{
"id": card_id,
"level": level.upper(),
"curriculum_order": curriculum_order,
"concept_id": _first(
record,
"concept_id",
"operation_id",
"composition_id",
"identity_id",
"motif_id",
"card_id",
"candidate_id",
),
"grammar_id": grammar_id,
"leakage_group_id": str(
record.get("leakage_group_id") or grammar_id
),
"variation_role": role,
"family": _family(level, record),
"label": _first(
record,
"label",
"identity_label",
"composition_label",
"operation_label",
"concept_label",
"mode_label",
),
"generator_program": (
prefix / files["program"].relative_to(bank)
).as_posix(),
"factor_assignment": (
read_json(factor_path) if factor_path is not None else {}
),
"parameters": (
read_json(parameter_path) if parameter_path is not None else {}
),
"deterministic_seed": None,
}
)
curriculum_order += 1
level_summaries[level] = {
"bank_name": bank_name,
"row_count": len(records),
"source_dataset_digest_sha256": source_digest,
"generator_version": str(manifest.get("generator_version") or ""),
"role_counts": dict(sorted(role_counts.items())),
}
inventory_entries = []
for relative, source in sorted(payload.items()):
source_bytes = _payload_bytes(source)
inventory_entries.append(
{
"path": relative,
"bytes": len(source_bytes),
"sha256": _sha256_bytes(source_bytes),
}
)
source_manifest = {
"schema_version": "pixcell-core-source-v1",
"core_name": CORE_NAME,
"source_commit": SOURCE_COMMIT,
"randomness": {
"mode": "none",
"seed": None,
"explanation": (
"core-v1 is an explicit deterministic enumeration; every factor "
"assignment and executable generator program is frozen."
),
},
"bank_names": BANK_NAMES,
"levels": level_summaries,
"file_count": len(inventory_entries),
"payload_bytes": sum(entry["bytes"] for entry in inventory_entries),
"files": inventory_entries,
}
grammar_catalog = {
"schema_version": "pixcell-grammar-catalog-v1",
"core_name": CORE_NAME,
"row_count": len(grammar_rows),
"representation_count": len({row["grammar_id"] for row in grammar_rows}),
"randomness": "none",
"rows": grammar_rows,
}
return payload, source_manifest, grammar_catalog
def _tar_info(name: str, size: int) -> tarfile.TarInfo:
info = tarfile.TarInfo(name=name)
info.size = size
info.mode = 0o644
info.uid = 0
info.gid = 0
info.uname = ""
info.gname = ""
info.mtime = 0
info.pax_headers = {}
return info
def write_deterministic_archive(
output_path: Path,
payload: dict[str, ARCHIVE_PAYLOAD],
source_manifest: dict[str, Any],
) -> None:
output_path.parent.mkdir(parents=True, exist_ok=True)
internal_manifest = (
json.dumps(
source_manifest,
indent=2,
sort_keys=True,
ensure_ascii=True,
allow_nan=False,
)
+ "\n"
).encode("utf-8")
with output_path.open("wb") as raw:
with gzip.GzipFile(
filename="",
mode="wb",
fileobj=raw,
compresslevel=9,
mtime=0,
) as compressed:
with tarfile.open(
mode="w",
fileobj=compressed,
format=tarfile.PAX_FORMAT,
) as archive:
for relative, source in sorted(payload.items()):
info = _tar_info(relative, _payload_size(source))
if isinstance(source, bytes):
archive.addfile(info, io.BytesIO(source))
else:
with source.open("rb") as handle:
archive.addfile(info, handle)
info = _tar_info("source_manifest.json", len(internal_manifest))
archive.addfile(info, io.BytesIO(internal_manifest))
def pack_core_v1(
source_root: Path,
dataset_root: Path,
) -> dict[str, Any]:
dataset_root = dataset_root.expanduser().resolve()
source_root = source_root.expanduser().resolve()
catalog = read_json(dataset_root / "metadata" / "catalog.json")
if catalog.get("release_digest_sha256") != EXPECTED_RELEASE_DIGEST:
raise ReleaseError("public release digest is not the frozen core-v1 digest")
if catalog.get("total_rows") != 738:
raise ReleaseError("public release row count is not 738")
payload, source_manifest, grammar_catalog = collect_source(source_root)
core_root = dataset_root / "factory" / CORE_NAME
archive_path = core_root / "source.tar.gz"
write_deterministic_archive(archive_path, payload, source_manifest)
write_json(core_root / "source_manifest.json", source_manifest)
write_json(core_root / "grammar_catalog.json", grammar_catalog)
archive_sha = sha256_file(archive_path)
admission_policy_path = dataset_root / "factory" / "admission_policy.json"
if not admission_policy_path.is_file():
raise ReleaseError("factory/admission_policy.json is missing")
reproducibility_report_path = (
dataset_root / "factory" / CORE_NAME / "reproducibility_report.json"
)
if not reproducibility_report_path.is_file():
raise ReleaseError("core-v1 reproducibility report is missing")
reference_paths = [
*(f"data/{level}/train-00000-of-00001.parquet" for level in LEVELS),
"galleries/all.png",
*(f"galleries/{level}.png" for level in LEVELS),
"metadata/catalog.json",
"metadata/audit_report.json",
*(f"metadata/source_reports/{level}.json" for level in LEVELS),
]
reference_artifacts = {
relative: sha256_file(dataset_root / relative) for relative in reference_paths
}
freeze = {
"schema_version": "pixcell-core-freeze-v1",
"core_name": CORE_NAME,
"release_digest_sha256": EXPECTED_RELEASE_DIGEST,
"row_count": 738,
"representation_count": 547,
"accepted_only": True,
"source_commit": SOURCE_COMMIT,
"levels": {
level: {
"row_count": EXPECTED_LEVELS[level]["rows"],
"source_dataset_digest_sha256": EXPECTED_LEVELS[level]["source_digest"],
}
for level in LEVELS
},
"source_bundle": {
"path": "factory/core-v1/source.tar.gz",
"sha256": archive_sha,
"bytes": archive_path.stat().st_size,
"inventory_path": "factory/core-v1/source_manifest.json",
"file_count": source_manifest["file_count"],
"uncompressed_payload_bytes": source_manifest["payload_bytes"],
},
"grammar_catalog": {
"path": "factory/core-v1/grammar_catalog.json",
"sha256": sha256_file(core_root / "grammar_catalog.json"),
"row_count": grammar_catalog["row_count"],
"representation_count": grammar_catalog["representation_count"],
},
"admission_policy": {
"path": "factory/admission_policy.json",
"sha256": sha256_file(admission_policy_path),
},
"reproducibility_report": {
"path": "factory/core-v1/reproducibility_report.json",
"sha256": sha256_file(reproducibility_report_path),
},
"determinism": {
"randomness": "none",
"seed": None,
"archive_metadata_normalized": True,
"expected_logical_digest_is_mandatory": True,
"reference_artifacts_are_platform_specific": True,
},
"reference_artifacts": reference_artifacts,
"build_command": "python scripts/build_all.py",
"validation_command": ("python scripts/validate_release.py --execute-programs"),
}
write_json(core_root / "freeze.json", freeze)
return freeze
def _parser() -> argparse.ArgumentParser:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument(
"--source",
type=Path,
required=True,
help="Working tree containing the five audited source banks.",
)
parser.add_argument(
"--dataset-root",
type=Path,
default=Path(__file__).resolve().parents[1],
help="Dataset package root (default: parent of scripts/).",
)
return parser
def main() -> None:
args = _parser().parse_args()
freeze = pack_core_v1(args.source, args.dataset_root)
bundle = freeze["source_bundle"]
print(
f"packed {freeze['row_count']} rows into {bundle['file_count']} source "
f"files; archive sha256 {bundle['sha256']}"
)
if __name__ == "__main__":
main()