File size: 9,626 Bytes
5ee3a5e | 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 | """통과한 P Formula student online과 5-output raster를 하나의 모바일 쌍으로 export한다."""
from __future__ import annotations
import argparse
import importlib.util
import json
from pathlib import Path
import sys
from typing import Any
import numpy as np
import torch
PROJECT_ROOT = Path(__file__).parents[1]
SOURCE_ROOT = PROJECT_ROOT / "src"
for path in (PROJECT_ROOT, SOURCE_ROOT):
if str(path) not in sys.path:
sys.path.insert(0, str(path))
from math_grid_drawer.research.ink06_canonical import render_canonical_ink
from math_grid_drawer.research.ink06_export import (
PFormulaStudentExportWrapper06,
RasterDebugExportWrapper06,
exported_equivalence06,
)
from math_grid_drawer.research.math_ink_06 import MathInk06Engine
from math_grid_drawer.research.p_formula_dataset06 import (
_formula_box06,
p_formula_symbol_ink06,
)
from math_grid_drawer.research.p_formula_gate06 import audit_p_formula_records06
from math_grid_drawer.research.skeleton_adapter06 import (
DualModalityTrajectoryAdapter06,
SkeletonTrajectoryAdapter06,
)
from scripts.export_math_ink_06_litert import (
_convert_litert,
_save_exported_program06,
_vocabulary_sha25606,
)
from scripts.export_math_ink_06_p_formula_student import (
_online_branch06,
_file_sha25606,
_read_jsonl06,
_resolve_checkpoint_path06,
validate_p_formula_student_artifacts06,
validate_p_formula_student_export06,
)
MAXIMUM_MODEL_BUNDLE_BYTES06 = 25 * 1024 * 1024
def paired_p_representatives06(
records: list[dict[str, Any]],
*,
maximum_samples: int,
) -> tuple[list[tuple[torch.Tensor, ...]], list[tuple[torch.Tensor, ...]]]:
"""필요 변수: P test formula·상한. 작동 원리: 같은 symbol에서 online 128×19와 raster 128×128을 함께 만든다."""
if maximum_samples <= 0:
raise ValueError("대표 입력 상한은 양수여야 합니다.")
online: list[tuple[torch.Tensor, ...]] = []
raster: list[tuple[torch.Tensor, ...]] = []
for record in records:
if str(record.get("split") or "") != "test":
continue
formula_box = _formula_box06(record)
for symbol in record["symbols"]:
ink = p_formula_symbol_ink06(symbol, formula_box=formula_box)
image = np.asarray(render_canonical_ink(ink), dtype=np.float32)
online.append((torch.from_numpy(ink.features).unsqueeze(0),))
raster.append((
torch.from_numpy(1.0 - image / 255.0).unsqueeze(0).unsqueeze(0),
))
if len(online) >= maximum_samples:
return online, raster
if not online:
raise ValueError("P Formula test representative가 없습니다.")
return online, raster
def _raster_branch06(adapter: torch.nn.Module) -> torch.nn.Module:
"""필요 변수: single/dual adapter. 작동 원리: image virtual stroke에 대응하는 raster branch만 반환한다."""
return adapter.raster if isinstance(adapter, DualModalityTrajectoryAdapter06) else adapter
def main() -> None:
"""필요 변수: 통과 student·동일 P corpus·출력. 작동 원리: 같은 lineage의 online/raster graph를 export한다."""
parser = argparse.ArgumentParser(description="Export Math Ink 0.6 P mobile pair")
parser.add_argument("--student-checkpoint", type=Path, required=True)
parser.add_argument("--base-checkpoint", type=Path)
parser.add_argument("--adapter-checkpoint", type=Path)
parser.add_argument("--data", type=Path, required=True)
parser.add_argument("--output", type=Path, required=True)
parser.add_argument("--maximum-representative-samples", type=int, default=256)
parser.add_argument("--convert-litert", action="store_true")
args = parser.parse_args()
payload = torch.load(args.student_checkpoint, map_location="cpu", weights_only=False)
data_sha256 = _file_sha25606(args.data)
validate_p_formula_student_export06(payload, data_sha256=data_sha256)
parent = args.student_checkpoint.parent
base = args.base_checkpoint or _resolve_checkpoint_path06(
payload["student_base_checkpoint"],
parent=parent,
)
adapter_checkpoint = args.adapter_checkpoint or _resolve_checkpoint_path06(
payload["student_online_adapter"],
parent=parent,
)
validate_p_formula_student_artifacts06(
payload,
base_checkpoint=base,
online_adapter=adapter_checkpoint,
)
engine = MathInk06Engine(base, adapter_checkpoint=adapter_checkpoint)
formula_adapter = SkeletonTrajectoryAdapter06(
hidden_size=int(payload["hidden_size"]),
)
formula_adapter.load_state_dict(payload["state_dict"])
online_wrapper = PFormulaStudentExportWrapper06(
engine.model,
_online_branch06(engine.composite_adapter),
formula_adapter,
family_weight=engine.online_family_fusion_weight,
exact_family_index=engine.exact_family_index,
).eval()
fusion = engine.raster_fusion
if any(float(fusion[key]) != 0.0 for key in ("family_weight", "geometry_weight", "symmetry_weight")):
raise ValueError("Mobile raster export는 학습 graph 밖 auxiliary fusion을 허용하지 않습니다.")
raster_wrapper = RasterDebugExportWrapper06(
engine.model,
adapter=_raster_branch06(engine.composite_adapter),
fusion_mode=str(fusion["mode"]),
score_weight=float(fusion["score_weight"]),
).eval()
records = _read_jsonl06(args.data)
audit = audit_p_formula_records06(records)
if not audit["eligible_for_product_evaluation"]:
raise ValueError("P Formula corpus가 product preflight를 통과하지 못했습니다.")
online_inputs, raster_inputs = paired_p_representatives06(
records,
maximum_samples=args.maximum_representative_samples,
)
online_export = torch.export.export(
online_wrapper,
online_inputs[0],
strict=True,
)
raster_export = torch.export.export(
raster_wrapper,
raster_inputs[0],
strict=True,
)
equivalence = {
"online": exported_equivalence06(
online_wrapper,
online_export,
online_inputs,
),
"raster": exported_equivalence06(
raster_wrapper,
raster_export,
raster_inputs,
),
}
args.output.mkdir(parents=True, exist_ok=True)
online_path = args.output / "p_formula_online.pt2"
raster_path = args.output / "raster_debug5.pt2"
_save_exported_program06(online_export, online_path)
_save_exported_program06(raster_export, raster_path)
total_bytes = online_path.stat().st_size + raster_path.stat().st_size
size_gate = total_bytes <= MAXIMUM_MODEL_BUNDLE_BYTES06
labels = tuple(str(label) for label in engine.labels)
report: dict[str, Any] = {
"schema": "aiflow-math-ink-06-p-mobile-pair-export-v1",
"model_version": f"{engine.model_version}+p-formula-student",
"student_checkpoint": str(args.student_checkpoint),
"data_sha256": data_sha256,
"teacher_seeds": [17, 31, 47],
"teacher_weights_embedded": False,
"exact_label_count": len(labels),
"vocabulary_sha256": _vocabulary_sha25606(list(labels)),
"raster_output_count": 5,
"representative_samples": len(online_inputs),
"preflight": audit,
"torch_version": torch.__version__,
"torch_export": {
"online": {
**equivalence["online"],
"path": online_path.name,
"bytes": online_path.stat().st_size,
},
"raster": {
**equivalence["raster"],
"path": raster_path.name,
"bytes": raster_path.stat().st_size,
},
"total_bytes": total_bytes,
"maximum_bundle_bytes": MAXIMUM_MODEL_BUNDLE_BYTES06,
"size_gate_passed": size_gate,
},
"torch_export_gate_passed": bool(
size_gate
and equivalence["online"]["gate_passed"]
and equivalence["raster"]["gate_passed"]
),
"litert_package_available": importlib.util.find_spec("litert_torch") is not None,
"litert": {
"online": {"converted": False, "reason": "conversion_not_requested"},
"raster": {"converted": False, "reason": "conversion_not_requested"},
},
"product_validation": False,
}
if args.convert_litert:
if not report["litert_package_available"]:
for branch in ("online", "raster"):
report["litert"][branch] = {
"converted": False,
"reason": "litert_torch_not_installed",
}
else:
report["litert"] = {
"online": _convert_litert(
online_wrapper,
online_inputs,
args.output / "p_formula_online.tflite",
),
"raster": _convert_litert(
raster_wrapper,
raster_inputs,
args.output / "raster_debug5.tflite",
),
}
report["next_gate"] = (
"package exact model pair, raster release validation, Android low/mid/high"
)
(args.output / "export_manifest.json").write_text(
json.dumps(report, ensure_ascii=False, indent=2) + "\n",
encoding="utf-8",
)
print(json.dumps(report, ensure_ascii=False, indent=2))
if __name__ == "__main__":
main()
|