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
|
#!/usr/bin/env python3
"""Apply the frozen A3 full seed-0 ResNet-20 advancement gate."""
import argparse
import json
import math
import os
SPLIT_HASH = "8328b206a97c420e49e54e3eca4abe3274c4756b084355784ea3fb8059e4515b"
def read_result(path, mode):
with open(path) as handle:
record = json.load(handle)
args = record["args"]
expected = {
"mode": mode, "depth": 20, "width": 16, "seed": 0,
"epochs": 200, "val_examples": 5000,
"eval_split": "validation", "normalization": "batchnorm",
}
for key, value in expected.items():
if args.get(key) != value:
raise ValueError(f"{path}: {key} drift")
if record["provenance"]["git_tracked_dirty"]:
raise ValueError(f"tracked-dirty result: {path}")
if record["split"]["validation_index_sha256"] != SPLIT_HASH:
raise ValueError(f"split drift: {path}")
if record["evaluation_protocol"]["test_evaluations"]:
raise ValueError(f"A3 touched test: {path}")
alignment = None
if record["diagnostics"] is not None:
alignment = float(record["diagnostics"]["early_third_mean"])
accuracy = float(record["final"]["accuracy"])
loss = float(record["final"]["loss"])
return {
"mode": mode, "path": path, "accuracy": accuracy, "loss": loss,
"finite": record["final"]["finite"] and math.isfinite(accuracy + loss),
"early_third_alignment": alignment,
"total_macs": record["work"]["total_macs_estimate"],
"peak_memory": record["hardware"]["peak_memory_allocated_bytes"],
"source_commit": record["provenance"]["git_commit"],
}
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--bp_selection", default="results/oral_a_bp_selection.json")
parser.add_argument("--dfa", default="results/oral_a_dev/dfa_full_r20_s0.json")
parser.add_argument("--sdil", default="results/oral_a_dev/sdil_full_r20_s0.json")
parser.add_argument("--out", default="results/oral_a_full_gate.json")
args = parser.parse_args()
with open(args.bp_selection) as handle:
bp_selection = json.load(handle)
if not bp_selection["status"].startswith("passed_"):
raise ValueError("A1 BP reference is not viable")
bp = read_result(bp_selection["selected"]["path"], "bp")
dfa = read_result(args.dfa, "dfa")
sdil = read_result(args.sdil, "sdil")
checks = {
"bp_at_least_90": bp["accuracy"] >= 0.90,
"sdil_within_5pt_bp": sdil["accuracy"] >= bp["accuracy"] - 0.05,
"sdil_beats_dfa_2pt": sdil["accuracy"] >= dfa["accuracy"] + 0.02,
"sdil_alignment_at_least_0p05": (
sdil["early_third_alignment"] is not None
and sdil["early_third_alignment"] >= 0.05),
"all_finite": all(row["finite"] for row in (bp, dfa, sdil)),
"sdil_macs_no_more_than_bp": sdil["total_macs"] <= bp["total_macs"],
}
output = {
"protocol": "oral_a_A3_v1",
"status": "passed" if all(checks.values()) else "failed",
"checks": checks, "rows": [bp, dfa, sdil],
"confirmation_test_seeds_touched": False,
}
os.makedirs(os.path.dirname(os.path.abspath(args.out)), exist_ok=True)
with open(args.out, "w") as handle:
json.dump(output, handle, indent=2, sort_keys=True)
handle.write("\n")
print(json.dumps(output, indent=2))
if __name__ == "__main__":
main()
|
