File size: 12,079 Bytes
12c2325 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 | #!/usr/bin/env python3
"""Recover the exact E4 alignment model source from the recorded forward patch."""
from __future__ import annotations
import argparse
import hashlib
import json
from pathlib import Path
JOURNAL_ROOT = Path("/nfs/ywang29/GmNet/journal_exp")
DEFAULT_OUTPUT = Path(
"/nfs/ywang29/GmNet/e4_spectral_audit/source_snapshots/alignment_4a62"
)
SESSION_PATH = Path(
"/root/.codex/sessions/2026/07/12/"
"rollout-2026-07-12T19-12-35-019f57be-e49b-77b3-9198-0721b55f06ff.jsonl"
)
ALIGNMENT_CODE_SHA256 = (
"4a62bc48370247d882196cd9afc1a0f1629a5bbfd522c3f599ffaf0c1187641f"
)
MECHANISM_CODE_SHA256 = (
"f6bd434a2ac86c772f2e23683431f3bb7ccb06f539558682a3bea6464bacebf9"
)
SOURCE_SPECS = {
"gmnet/models/gmnet.py": {
"current_sha256": (
"8446a0ed7d6863f81e88929f3277d363c961372210279de7d0e770f636852f3c"
),
"recovered_sha256": (
"5291156084aaf72936d1f8896057e7e606f74ef05aed856df4940e4838dbe3ac"
),
"recovered_size": 28103,
"session_events": [
{
"timestamp": "2026-07-17T21:00:58.920Z",
"jsonl_line": 6665,
},
{
"timestamp": "2026-07-17T21:01:05.385Z",
"jsonl_line": 6670,
},
{
"timestamp": "2026-07-17T21:03:01.840Z",
"jsonl_line": 6723,
},
],
},
"gmnet/models/__init__.py": {
"current_sha256": (
"68a705c94c1223305f2d5ab18cc41faf36c35d90748fde5a32f4474222e4ef29"
),
"recovered_sha256": (
"9227138f9cd0e0110f57bad21b9302d8b5daa51cd63246e37c8ff2878eb2715e"
),
"recovered_size": 722,
"session_events": [
{
"timestamp": "2026-07-17T21:03:01.840Z",
"jsonl_line": 6723,
}
],
},
}
GMNET_REVERSE_REPLACEMENTS = (
(
'''SUPPORTED_GATE_INTERVENTIONS: Final[tuple[str, ...]] = (\n'''
''' "baseline",\n'''
''' "stop_gradient",\n'''
''' "channel_derangement",\n'''
''' "batch_derangement",\n'''
''' "stop_gradient_channel_derangement",\n'''
''')\n'''
'''CHANNEL_DERANGEMENT_INTERVENTIONS: Final[frozenset[str]] = frozenset(\n'''
''' {"channel_derangement", "stop_gradient_channel_derangement"}\n'''
''')\n'''
'''STOP_GRADIENT_INTERVENTIONS: Final[frozenset[str]] = frozenset(\n'''
''' {"stop_gradient", "stop_gradient_channel_derangement"}\n'''
''')\n''',
'''SUPPORTED_GATE_INTERVENTIONS: Final[tuple[str, ...]] = (\n'''
''' "baseline",\n'''
''' "stop_gradient",\n'''
''' "channel_derangement",\n'''
''')\n''',
),
(
''' permutation = (\n'''
''' _channel_derangement(channels, seed)\n'''
''' if intervention in CHANNEL_DERANGEMENT_INTERVENTIONS\n'''
''' else None\n'''
''' )\n''',
''' permutation = (\n'''
''' _channel_derangement(channels, seed)\n'''
''' if intervention == "channel_derangement"\n'''
''' else None\n'''
''' )\n''',
),
(
''' value = x\n'''
''' if self.gate_intervention == "batch_derangement":\n'''
''' if x.ndim < 1 or x.shape[0] < 2:\n'''
''' raise ValueError(\n'''
''' "batch_derangement requires at least two samples per local "\n'''
''' f"batch, got shape {tuple(x.shape)}"\n'''
''' )\n'''
''' shift = self.batch_derangement_shift(int(x.shape[0]))\n'''
''' value = x.roll(shifts=-shift, dims=0)\n'''
''' elif self.gate_intervention in CHANNEL_DERANGEMENT_INTERVENTIONS:\n'''
''' if x.ndim < 2 or x.shape[1] != self._gate_channels:\n'''
''' raise ValueError(\n'''
''' "channel derangement expected channel dimension "\n'''
''' f"{self._gate_channels}, got shape {tuple(x.shape)}"\n'''
''' )\n'''
''' assert self.gate_permutation is not None\n'''
''' value = x.index_select(1, self.gate_permutation)\n'''
''' if self.gate_intervention in STOP_GRADIENT_INTERVENTIONS:\n'''
''' value = value.detach()\n'''
''' return value\n'''
'''\n'''
''' def batch_derangement_shift(self, batch_size: int) -> int:\n'''
''' """Return the seeded cyclic shift used for a local batch derangement."""\n'''
'''\n'''
''' if isinstance(batch_size, bool) or not isinstance(batch_size, int):\n'''
''' raise TypeError("batch_size must be an integer")\n'''
''' if batch_size < 2:\n'''
''' raise ValueError("batch derangement requires batch_size >= 2")\n'''
''' return 1 + self.gate_intervention_seed % (batch_size - 1)\n''',
''' if self.gate_intervention == "baseline":\n'''
''' return x\n'''
''' if self.gate_intervention == "stop_gradient":\n'''
''' return x.detach()\n'''
''' if x.ndim < 2 or x.shape[1] != self._gate_channels:\n'''
''' raise ValueError(\n'''
''' "channel_derangement expected channel dimension "\n'''
''' f"{self._gate_channels}, got shape {tuple(x.shape)}"\n'''
''' )\n'''
''' assert self.gate_permutation is not None\n'''
''' return x.index_select(1, self.gate_permutation)\n''',
),
(
''' "is_bijection": is_bijection,\n'''
''' "stops_gate_gradient": (\n'''
''' module.gate_intervention\n'''
''' in STOP_GRADIENT_INTERVENTIONS\n'''
''' ),\n'''
''' "channel_derangement": (\n'''
''' module.gate_intervention\n'''
''' in CHANNEL_DERANGEMENT_INTERVENTIONS\n'''
''' ),\n'''
''' "batch_derangement": (\n'''
''' module.gate_intervention == "batch_derangement"\n'''
''' ),\n'''
''' "batch_shift_rule": (\n'''
''' "1 + seed % (local_batch_size - 1)"\n'''
''' if module.gate_intervention == "batch_derangement"\n'''
''' else None\n'''
''' ),\n''',
''' "is_bijection": is_bijection,\n''',
),
(
'''__all__ = [\n "CHANNEL_DERANGEMENT_INTERVENTIONS",\n''',
'''__all__ = [\n''',
),
(
''' "SmoothClippedSelfGate",\n "STOP_GRADIENT_INTERVENTIONS",\n''',
''' "SmoothClippedSelfGate",\n''',
),
)
INIT_REVERSE_REPLACEMENTS = (
(
'''from .gmnet import (\n CHANNEL_DERANGEMENT_INTERVENTIONS,\n''',
'''from .gmnet import (\n''',
),
(
''' SmoothClippedSelfGate,\n STOP_GRADIENT_INTERVENTIONS,\n''',
''' SmoothClippedSelfGate,\n''',
),
(
'''__all__ = [\n "CHANNEL_DERANGEMENT_INTERVENTIONS",\n''',
'''__all__ = [\n''',
),
(
''' "SmoothClippedSelfGate",\n "STOP_GRADIENT_INTERVENTIONS",\n''',
''' "SmoothClippedSelfGate",\n''',
),
)
def sha256_bytes(value: bytes) -> str:
return hashlib.sha256(value).hexdigest()
def recover(relative_path: str, replacements: tuple[tuple[str, str], ...]) -> bytes:
spec = SOURCE_SPECS[relative_path]
source = (JOURNAL_ROOT / relative_path).read_bytes()
actual = sha256_bytes(source)
if actual != spec["current_sha256"]:
raise RuntimeError(
f"current {relative_path} SHA256 mismatch: {actual} != "
f"{spec['current_sha256']}"
)
text = source.decode("utf-8")
for index, (postimage, preimage) in enumerate(replacements):
count = text.count(postimage)
if count != 1:
raise RuntimeError(
f"reverse replacement {index} for {relative_path} matched {count} times"
)
text = text.replace(postimage, preimage)
recovered = text.encode("utf-8")
if len(recovered) != spec["recovered_size"]:
raise RuntimeError(
f"recovered {relative_path} size mismatch: {len(recovered)} != "
f"{spec['recovered_size']}"
)
digest = sha256_bytes(recovered)
if digest != spec["recovered_sha256"]:
raise RuntimeError(
f"recovered {relative_path} SHA256 mismatch: {digest} != "
f"{spec['recovered_sha256']}"
)
return recovered
def build_manifest(files: dict[str, bytes]) -> dict[str, object]:
return {
"schema_version": 1,
"snapshot_id": "alignment_4a62_model_source",
"alignment_code_sha256": ALIGNMENT_CODE_SHA256,
"source_postimage_code_sha256": MECHANISM_CODE_SHA256,
"recovery_method": "exact_reverse_of_recorded_forward_apply_patch",
"session_path": str(SESSION_PATH),
"files": [
{
"path": path,
"size": len(content),
"sha256": sha256_bytes(content),
"alignment_manifest_sha256": SOURCE_SPECS[path][
"recovered_sha256"
],
"session_events": SOURCE_SPECS[path]["session_events"],
}
for path, content in sorted(files.items())
],
}
def main() -> int:
parser = argparse.ArgumentParser()
parser.add_argument("--output", type=Path, default=DEFAULT_OUTPUT)
parser.add_argument("--check", action="store_true")
args = parser.parse_args()
# The original session is required once, at recovery time. Formal launch
# containers only validate the already frozen snapshot against the encoded
# reverse patch and hashes, so they do not depend on /root session storage.
if not args.check and not SESSION_PATH.is_file():
raise FileNotFoundError(f"recorded Codex session is missing: {SESSION_PATH}")
files = {
"gmnet/models/gmnet.py": recover(
"gmnet/models/gmnet.py", GMNET_REVERSE_REPLACEMENTS
),
"gmnet/models/__init__.py": recover(
"gmnet/models/__init__.py", INIT_REVERSE_REPLACEMENTS
),
}
manifest = build_manifest(files)
manifest_bytes = (
json.dumps(manifest, indent=2, sort_keys=True) + "\n"
).encode("utf-8")
if args.check:
for relative_path, expected in files.items():
path = args.output / relative_path
if not path.is_file() or path.read_bytes() != expected:
raise RuntimeError(f"snapshot mismatch: {path}")
manifest_path = args.output / "source_manifest.json"
if not manifest_path.is_file() or manifest_path.read_bytes() != manifest_bytes:
raise RuntimeError(f"snapshot manifest mismatch: {manifest_path}")
print(f"Validated exact alignment model snapshot at {args.output}")
return 0
for relative_path, content in files.items():
destination = args.output / relative_path
destination.parent.mkdir(parents=True, exist_ok=True)
destination.write_bytes(content)
(args.output / "source_manifest.json").write_bytes(manifest_bytes)
print(f"Recovered exact alignment model snapshot at {args.output}")
return 0
if __name__ == "__main__":
raise SystemExit(main())
|