Publish fixed regional-context item b entry and adopted item1b review with capacity hold
This commit is contained in:
1 parent
ffd5c92958
commit
31a20abd43
34 files changed
+16026
No files matched your search
@@ -0,0 +1,75 @@
|
||||
"""Independent 1b entry review: opaque bindings, metadata, AST and CPU tests."""
|
||||
import ast,copy,hashlib,json,os,resource,subprocess,sys,time
|
||||
from pathlib import Path
|
||||
|
||||
B=Path('/home/frank14f/optane/MeltMind')
|
||||
R=Path('/home/frank14f/MeltMind-runs/phase3-item-1b-2026-11-18')
|
||||
OUT=Path(__file__).resolve().parent
|
||||
def read(p):return json.loads(Path(p).read_text())
|
||||
def sha(p):
|
||||
h=hashlib.sha256()
|
||||
with Path(p).open('rb') as f:
|
||||
for b in iter(lambda:f.read(1048576),b''):h.update(b)
|
||||
return h.hexdigest()
|
||||
def write(n,d):
|
||||
with (OUT/n).open('x') as f:json.dump(d,f,indent=2);f.write('\n')
|
||||
c=read(R/'contract.json');p=read(R/'implementation-git-freeze.json');binding=read(R/'expected-review-binding.json')
|
||||
expected=dict(reviewed_commit=p['commit'],contract_sha256=sha(R/'contract.json'),implementation_proof_sha256=sha(R/'implementation-git-freeze.json'),comparison_contract_sha256=sha(R/'comparison-contract.json'),code_sha256=c['code_sha256'])
|
||||
assert binding==expected
|
||||
assert p['commit']=='45b5df30098f74dc3c93a2f54365d7d866b810e0'
|
||||
assert expected['contract_sha256']=='d33f036adc48cde904fbb6bb0f615a5464f94b039906d0b45a4dbd15f0681bd3'
|
||||
assert expected['implementation_proof_sha256']=='73f28ef6d9991ef36f9e399b39ff24d849a656ee3caa8b3230e3c61732839f29'
|
||||
assert expected['comparison_contract_sha256']=='648a99549689ae6aba02085d59fc5b9a10d350e48832c383570569a657d24aae'
|
||||
assert sha(R/'expected-review-binding.json')=='13a5142ee9d1d635b9d7f2979c2cd392de8feda8fcca722fcdc469a2d9927873'
|
||||
for mapping in [c['code_sha256'],c['source_sha256'],p['files']]:
|
||||
for path,h in mapping.items():assert sha(path)==h,path
|
||||
for path,h in p['files'].items():
|
||||
raw=subprocess.check_output(['git','-C',p['local_repository'],'show',p['commit']+':'+p['git_paths'][path]])
|
||||
assert hashlib.sha256(raw).hexdigest()==h,path
|
||||
from phase3_item1b_common import config
|
||||
config(R)
|
||||
old_pins={}
|
||||
for folder in ['phase3-item-c-2026-11-18','phase3-seed1-v2-2026-11-18','track-a-recovery-2026-11-18']:
|
||||
old=read(R.parent/folder/'contract.json')
|
||||
for path,h in old['code_sha256'].items():assert sha(path)==h,path
|
||||
old_pins[folder]=len(old['code_sha256'])
|
||||
reviewed_shared=read(B/'artifacts/meltassist-item-c-reservation-delta-review-2026-11-18/review.json')
|
||||
shared_path=str(B/'tools/phase3_shared_capacity.py')
|
||||
assert c['code_sha256'][shared_path]==reviewed_shared['code_sha256'][shared_path]
|
||||
|
||||
def epoch_loop(name):
|
||||
tree=ast.parse((B/'tools'/name).read_text())
|
||||
return next(n for n in ast.walk(tree) if isinstance(n,ast.For) and ast.unparse(n.target)=='epoch')
|
||||
oldloop=copy.deepcopy(epoch_loop('phase2_h2_s2_worker.py'))
|
||||
oldloop.iter.args[1].value=101
|
||||
for node in ast.walk(oldloop):
|
||||
if isinstance(node,ast.Constant) and node.value=='WAIT_ALL_FIRST_EPOCH_COST_GATE':node.value='WAIT_THIS_FOLD_FIRST_EPOCH_COST_GATE'
|
||||
if isinstance(node,ast.If) and ast.unparse(node.test)=='epoch < 200':node.test.comparators[0].value=100
|
||||
if isinstance(node,ast.Call) and isinstance(node.func,ast.Name) and node.func.id=='first_epoch_release':node.args=[ast.Name(id='root',ctx=ast.Load()),ast.Name(id='fold_id',ctx=ast.Load())]
|
||||
assert ast.dump(oldloop)==ast.dump(epoch_loop('phase3_item1b_worker.py'))
|
||||
|
||||
# Historical receipts are used only for model identity and validation timing,
|
||||
# never historical metric values or target/prediction/model array values.
|
||||
paired=[]
|
||||
for model,job in zip(c['baseline_models'],c['jobs']):
|
||||
assert model['fold']==job['id'] and model['completed_epochs']==200 and model['seed']==0
|
||||
result=read(model['result']);stats=read(model['standardization'])
|
||||
assert result['completed_epochs']==200 and result['optimizer_steps']==200*len(job['learning'])
|
||||
assert result['checkpoint']==model['checkpoint'] and result['checkpoint_sha256']==model['checkpoint_sha256']
|
||||
assert stats['parts']==job['learning'] and stats['nodes']==job['learning_nodes']
|
||||
hist=Path(model['result']).parent/'history'
|
||||
bound=max(read(hist/f'{e:03d}.json')['validation_seconds'] for e in range(10,201,10))
|
||||
assert bound==job['validation_seconds_bound']
|
||||
paired.append(dict(fold=job['id'],checkpoint_sha256=model['checkpoint_sha256'],validation_seconds_bound=bound,target_steps=job['target_steps']))
|
||||
|
||||
env=dict(os.environ,PYTHONDONTWRITEBYTECODE='1',CUDA_VISIBLE_DEVICES='',PYTHONPATH=str(B/'tools'),OPENBLAS_NUM_THREADS='1',OMP_NUM_THREADS='1',MKL_NUM_THREADS='1')
|
||||
u0=resource.getrusage(resource.RUSAGE_CHILDREN);start=time.monotonic()
|
||||
with (OUT/'synthetic-tests.log').open('x') as f:
|
||||
run=subprocess.run([sys.executable,'-m','unittest','-v','tools/test_phase3_item1b.py'],cwd=B,env=env,stdout=f,stderr=subprocess.STDOUT,timeout=60)
|
||||
u=resource.getrusage(resource.RUSAGE_CHILDREN)
|
||||
tests=dict(exit_code=run.returncode,wall_seconds=time.monotonic()-start,child_cpu_seconds=u.ru_utime+u.ru_stime-u0.ru_utime-u0.ru_stime,log_sha256=sha(OUT/'synthetic-tests.log'))
|
||||
assert run.returncode==0,tests
|
||||
assert 'Ran 23 tests' in (OUT/'synthetic-tests.log').read_text()
|
||||
evidence=dict(status='PASS',binding=expected,binding_file_sha256=sha(R/'expected-review-binding.json'),code_files=len(c['code_sha256']),source_files=len(c['source_sha256']),local_git_files=len(p['files']),old_code_pins=old_pins,shared_code_sha256=c['code_sha256'][shared_path],training_and_validation_loop_ast='Equal to original S2 after declared100epoch limit and per-fold cost-gate substitutions; learning-rate function independently tested',paired_baseline_models=paired,tests=tests,entry_state={n:(R/n).exists() for n in ['review.json','started.json','complete.json']},scientific_execution=False,remote_raw_bytes_independently_checked=False)
|
||||
write('evidence.json',evidence)
|
||||
print(json.dumps({k:v for k,v in evidence.items() if k not in ['binding','paired_baseline_models']}))
|
||||
@@ -0,0 +1,127 @@
|
||||
{
|
||||
"status": "PASS",
|
||||
"binding": {
|
||||
"reviewed_commit": "45b5df30098f74dc3c93a2f54365d7d866b810e0",
|
||||
"contract_sha256": "d33f036adc48cde904fbb6bb0f615a5464f94b039906d0b45a4dbd15f0681bd3",
|
||||
"implementation_proof_sha256": "73f28ef6d9991ef36f9e399b39ff24d849a656ee3caa8b3230e3c61732839f29",
|
||||
"comparison_contract_sha256": "648a99549689ae6aba02085d59fc5b9a10d350e48832c383570569a657d24aae",
|
||||
"code_sha256": {
|
||||
"/home/frank14f/optane/MeltMind/tools/freeze_g1_components.py": "d0cee60d95243ac2c59ef3f2244ba818315718c4e3d1895f2f9f064f1f7f5ecd",
|
||||
"/home/frank14f/optane/MeltMind/tools/gre24_models.py": "53b6fee8b700b9b982b06d29908ce2ce23e0c57512c46dbf74b2aade79e3ac3a",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_review.py": "915617aeb8292d03c654e637bc34a014c254fd178a773f135d1806c4b41372a8",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h1_worker_v2.py": "1543df85b615bb69b79bacd572074a7c77ae19dab110197601573a7e5805ed20",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_data.py": "7e03dd7a94e7bb43593277f183c222473d7cb7aaf2f60e53b45ab8d9499fcb09",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_model.py": "fab2470dd0654ca74c3b7d893f35ad1f3ae6c5c08ffe334ef6312cf6ec87892a",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_pilot_worker.py": "14651ca402e6d02466d8afdb77bc9e62fb2e752f0c0c3d20d98593157d5ebe81",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_home_freeze_io.py": "bf09204b85ed2a7038acdd202257f618474d1110a9d6d772f105a3ffb334d3ff",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_local_freeze_io.py": "60adf3382577613c5234f5cf30779336b9951b66e90c075cb7cd0adab9cc42dd",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_residual_math.py": "ffbd9624dec79bdaabf95244fe87209055908f31a977e73ea4377045d6f8fe62",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_capacity.py": "771faad0095712fd4fc8131ddcfabaa3901141592d2b055ea2faadf8ea3d3edf",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_common.py": "4e25effe8326ff8c7c3f751590315040209af054b32b0505dc3c6b001589db32",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_guard.py": "08b7d8ab2ac773fab8b749f9f61af259254422128e9634f8b715c81efce2651f",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_queue.py": "31dfd71369c63512c1d6fe176fd22e8db646520c92da533bb5530ca3843c388d",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_schedule.py": "06cb97594c3a87f42f712dcde6cec8c2d90e0b50a89a6be7954bdd2f5c18d5de",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_worker.py": "b781e56c519401bd38af6a522deb24506cba6993d2af008f85ea4e1715f1d9c6",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_seed1_v2_common.py": "25ca6be14027369393bd3616209b08f30e25892cccca83300483c5a116e8d75d",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_shared_capacity.py": "7c07b0aaf1bc4e6a980d016defcac326c3bd871134d2583bb4dcd88bb34728f8",
|
||||
"/home/frank14f/optane/MeltMind/tools/prepare_phase3_item1b.py": "c7bc9f3b7cf4137f52e1efad19dcb7045282d4e9a9c1aae50f066c9e488f7c01",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_gpu_idle_gate.py": "e2b1613e7e6ea398a8d555220a9365a1ad998ab4ef4b2bf8ec7cc4e31f8b4b9f",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_main_memory_guard.py": "e094f5443e11518efe9018215c11718a8e18aa462a977bf43e974b2d0b2aebb1",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/__init__.py": "a0fa1323d3553172da66c5d442c5fc4fcb0d5055c886c776928604550ea97609",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/attention.py": "7714adfccf6a0fc6b52a44aee3a50321a88898bf3d743680a1108c01e1ff8a61",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/blocks.py": "b13e986cea5f1d7f325a1638c1505197b4a105baacff5a9a515f23fcaf28ea75",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/features.py": "3b34ae61cf8012ebd58b7505a73d7aa43d8ee1890ac54f98d7917b67887a45a4",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/network.py": "5137e845d81d1223afc13bf87e03ccba1132a3baa75d5189a8f320b2752be071",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/normalization.py": "7180696d3845799e28d8c31ddef957d075de6a2e46d96a3a0f0df16fc05234a0",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/objectives.py": "5b6527a45f08690055d572335470ce5979e8bb86e1cab00de594f79ae21316b1",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_pilot_train.py": "e9cceead51a5cc092aad23409ad98909c84a62e7e10571e750d1e9be1ad3b948",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_precision_continuation_guard.py": "01235d897c7c03e5ffa109c58b14f5d24d128e8f1ba873e1f76cfb14e02f7a7f",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_recovery_gpu_supervisor.py": "689c0f3179e4fcd362e546d90214829a9e4f8739abdbdf74b560763e5a392f15",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_recovery_monitoring.py": "442c2a9c75e8c00b30a8f2e91cb52d87c466b19a80aea9128b10c021e61b922e",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase3_item1b.py": "b4d78f4384651711fe83ff00eb6719c6c61664699d801ecb34978b9546439da9",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_h2_common.py": "79918e7de887d1f0a2de5523156b27d00dc2d85ac930a07efc34d817d904fe65",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_h2_prepare.py": "f657be036dc48fc5936f56c4b8510a4b295d4787af6877cc5ccaad663107be17",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_recovery_monitor.py": "4c423f8b0dff69304d0046c0a9c319bb31aa0c6d1cbaeb3942726b5e6caa0496",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_recovery_supervisor.py": "5ccb2a6eec95b0700b2063a2f8df414037528e90c206648c1c5e49c5e89e741d",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_common.py": "38a29a34663a5d6310571eeff53b5d18a0355e51b9fa58a08033f64fa379a9df",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_common.py": "4e85f5d211c4d3e2188422aa4c0cefa40dcfa14ef32e90894ed3412de1fd1d12",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_data.py": "fe3aa3e023f22f4fe55cd47e7ffa677b58f52a041867ac05620cbd329d913cb6",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_guard.py": "7144fa77ffe61a2a3c92c050234e370ef4dfeb0e31fafb419f752f969737ff1a",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_loss.py": "79e291866bda300ecdc5577c62903d34abeec14f5cc36db05e346f9ffd64c24b",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_queue.py": "a04879cc0169ecb523a88360cddca1e234d91b3adbc20b3ec31d02461ed767eb",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_worker.py": "0f8f35e1422021b2a4f097c2e62267fb935fe4658fd0ee8ccd251609a4062ea7",
|
||||
"/home/frank14f/optane/MeltMind/tools/prepare_phase3_item_c.py": "782c415b1f536eb2b85aa2c9c4a9a0dc278f9b7134e8a83e2a6af8199396c53a",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_custody_common.py": "588c646101be0d50f505f9a12a1c98361aa3437c8c594eecb7321831f717a741",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_custody_statistics.py": "188565801f6901d0192f79aec0889e632501e672a4149d76d1835efbb9b45447",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase3_item_c.py": "3115aa9049974eac86657aa4cee7b5d716f6ee93268555cc86242029ebefce4c",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase2_h2_s2.py": "4235890fd194b7778e937f0be9245b964512df2c75d962ef703e3a988c5b8805",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_guard.py": "6bc7624b0cfe850591f77153cd2034eefdd62f7e63b93211ccdb87eb8aba346c",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_common.py": "308486507ae3589a0e6cc40565f6e60e6d2f63ae74c4d828bdacf33b3b8b8062",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_fixed_io.py": "ea17fdfc7508f3f980e3cdf69534b1eb8ceb8a8e469c32451b2d5e16d0ada575",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_worker.py": "37dfd75fff50b3d65f4ef37f4255536616582fe5f3d01155a0e037325ab5c7cd",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_worker.py": "ecee04b183ef2b93b7b39a447d48c40a5843d3190e48d2d7eb4939811a5db9b0",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_prepare.py": "964fae13ac95e4de4dc50b3c0a80791d86349ce3a9f3c5a6e7dc3cf8ca041fad",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_prepare.py": "3a31af0766f628a80f38e0b7286beb11ca3109188eabc4adfdb2ecb0eef3dc27",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_queue.py": "d82dfb365956e08250758bb4c651a19fbecce92fe209072f5f8ed35970450735",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_launch.py": "480fac25dea5a2006bb84130a4b0a0b253b3040b2150a56c38eada62b2c259e6",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_guard.py": "ae94711adc17a7630f398e8d4120fa3cc8f60979a00122c4cd2dc364c382ced5",
|
||||
"/home/frank14f/optane/MeltMind/docs/phase2-h2-s2-execution-contract-2026-11-17.md": "1e8e5ef31650519737cd6cd247abc0bd04744dd93ffb013eff88866d42f03073"
|
||||
}
|
||||
},
|
||||
"binding_file_sha256": "13a5142ee9d1d635b9d7f2979c2cd392de8feda8fcca722fcdc469a2d9927873",
|
||||
"code_files": 60,
|
||||
"source_files": 188,
|
||||
"local_git_files": 74,
|
||||
"old_code_pins": {
|
||||
"phase3-item-c-2026-11-18": 52,
|
||||
"phase3-seed1-v2-2026-11-18": 51,
|
||||
"track-a-recovery-2026-11-18": 27
|
||||
},
|
||||
"shared_code_sha256": "7c07b0aaf1bc4e6a980d016defcac326c3bd871134d2583bb4dcd88bb34728f8",
|
||||
"training_and_validation_loop_ast": "Equal to original S2 after declared100epoch limit and per-fold cost-gate substitutions; learning-rate function independently tested",
|
||||
"paired_baseline_models": [
|
||||
{
|
||||
"fold": 0,
|
||||
"checkpoint_sha256": "35cc9e41a4197e186570089ac305c4dee100f3d37611e0ca17e01db441b93756",
|
||||
"validation_seconds_bound": 12.383006267948076,
|
||||
"target_steps": 32000
|
||||
},
|
||||
{
|
||||
"fold": 1,
|
||||
"checkpoint_sha256": "8c457d5fb707707059cef76761d84ff9956bf45073845c9c486360f95a18b46e",
|
||||
"validation_seconds_bound": 11.502408961998299,
|
||||
"target_steps": 37400
|
||||
},
|
||||
{
|
||||
"fold": 2,
|
||||
"checkpoint_sha256": "f2bd3248352486874d94d67492159d77bf42b9fbd192969f0c603543847ded05",
|
||||
"validation_seconds_bound": 9.668979812064208,
|
||||
"target_steps": 36600
|
||||
},
|
||||
{
|
||||
"fold": 3,
|
||||
"checkpoint_sha256": "0c9137d2e88e4f53171a66b3a34811d541ddf781baad6612a1091b453b9b6a47",
|
||||
"validation_seconds_bound": 14.438761183992028,
|
||||
"target_steps": 29100
|
||||
},
|
||||
{
|
||||
"fold": 4,
|
||||
"checkpoint_sha256": "caf6d489a0842a2f705b70e70fcb0c14c41cbf2772d6adc006979f3a3f070861",
|
||||
"validation_seconds_bound": 7.502535316976719,
|
||||
"target_steps": 35000
|
||||
}
|
||||
],
|
||||
"tests": {
|
||||
"exit_code": 0,
|
||||
"wall_seconds": 4.178926190012135,
|
||||
"child_cpu_seconds": 4.113682,
|
||||
"log_sha256": "ff7715d22c8c0594fe184ae8da0a52ac5798f6651138d4ebe93493745bc385ab"
|
||||
},
|
||||
"entry_state": {
|
||||
"review.json": false,
|
||||
"started.json": false,
|
||||
"complete.json": false
|
||||
},
|
||||
"scientific_execution": false,
|
||||
"remote_raw_bytes_independently_checked": false
|
||||
}
|
||||
@@ -0,0 +1,95 @@
|
||||
{
|
||||
"reviewed_commit": "45b5df30098f74dc3c93a2f54365d7d866b810e0",
|
||||
"contract_sha256": "d33f036adc48cde904fbb6bb0f615a5464f94b039906d0b45a4dbd15f0681bd3",
|
||||
"implementation_proof_sha256": "73f28ef6d9991ef36f9e399b39ff24d849a656ee3caa8b3230e3c61732839f29",
|
||||
"comparison_contract_sha256": "648a99549689ae6aba02085d59fc5b9a10d350e48832c383570569a657d24aae",
|
||||
"code_sha256": {
|
||||
"/home/frank14f/optane/MeltMind/tools/freeze_g1_components.py": "d0cee60d95243ac2c59ef3f2244ba818315718c4e3d1895f2f9f064f1f7f5ecd",
|
||||
"/home/frank14f/optane/MeltMind/tools/gre24_models.py": "53b6fee8b700b9b982b06d29908ce2ce23e0c57512c46dbf74b2aade79e3ac3a",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_review.py": "915617aeb8292d03c654e637bc34a014c254fd178a773f135d1806c4b41372a8",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h1_worker_v2.py": "1543df85b615bb69b79bacd572074a7c77ae19dab110197601573a7e5805ed20",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_data.py": "7e03dd7a94e7bb43593277f183c222473d7cb7aaf2f60e53b45ab8d9499fcb09",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_model.py": "fab2470dd0654ca74c3b7d893f35ad1f3ae6c5c08ffe334ef6312cf6ec87892a",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_pilot_worker.py": "14651ca402e6d02466d8afdb77bc9e62fb2e752f0c0c3d20d98593157d5ebe81",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_home_freeze_io.py": "bf09204b85ed2a7038acdd202257f618474d1110a9d6d772f105a3ffb334d3ff",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_local_freeze_io.py": "60adf3382577613c5234f5cf30779336b9951b66e90c075cb7cd0adab9cc42dd",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_residual_math.py": "ffbd9624dec79bdaabf95244fe87209055908f31a977e73ea4377045d6f8fe62",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_capacity.py": "771faad0095712fd4fc8131ddcfabaa3901141592d2b055ea2faadf8ea3d3edf",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_common.py": "4e25effe8326ff8c7c3f751590315040209af054b32b0505dc3c6b001589db32",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_guard.py": "08b7d8ab2ac773fab8b749f9f61af259254422128e9634f8b715c81efce2651f",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_queue.py": "31dfd71369c63512c1d6fe176fd22e8db646520c92da533bb5530ca3843c388d",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_schedule.py": "06cb97594c3a87f42f712dcde6cec8c2d90e0b50a89a6be7954bdd2f5c18d5de",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_worker.py": "b781e56c519401bd38af6a522deb24506cba6993d2af008f85ea4e1715f1d9c6",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_seed1_v2_common.py": "25ca6be14027369393bd3616209b08f30e25892cccca83300483c5a116e8d75d",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_shared_capacity.py": "7c07b0aaf1bc4e6a980d016defcac326c3bd871134d2583bb4dcd88bb34728f8",
|
||||
"/home/frank14f/optane/MeltMind/tools/prepare_phase3_item1b.py": "c7bc9f3b7cf4137f52e1efad19dcb7045282d4e9a9c1aae50f066c9e488f7c01",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_gpu_idle_gate.py": "e2b1613e7e6ea398a8d555220a9365a1ad998ab4ef4b2bf8ec7cc4e31f8b4b9f",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_main_memory_guard.py": "e094f5443e11518efe9018215c11718a8e18aa462a977bf43e974b2d0b2aebb1",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/__init__.py": "a0fa1323d3553172da66c5d442c5fc4fcb0d5055c886c776928604550ea97609",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/attention.py": "7714adfccf6a0fc6b52a44aee3a50321a88898bf3d743680a1108c01e1ff8a61",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/blocks.py": "b13e986cea5f1d7f325a1638c1505197b4a105baacff5a9a515f23fcaf28ea75",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/features.py": "3b34ae61cf8012ebd58b7505a73d7aa43d8ee1890ac54f98d7917b67887a45a4",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/network.py": "5137e845d81d1223afc13bf87e03ccba1132a3baa75d5189a8f320b2752be071",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/normalization.py": "7180696d3845799e28d8c31ddef957d075de6a2e46d96a3a0f0df16fc05234a0",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/objectives.py": "5b6527a45f08690055d572335470ce5979e8bb86e1cab00de594f79ae21316b1",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_pilot_train.py": "e9cceead51a5cc092aad23409ad98909c84a62e7e10571e750d1e9be1ad3b948",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_precision_continuation_guard.py": "01235d897c7c03e5ffa109c58b14f5d24d128e8f1ba873e1f76cfb14e02f7a7f",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_recovery_gpu_supervisor.py": "689c0f3179e4fcd362e546d90214829a9e4f8739abdbdf74b560763e5a392f15",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_recovery_monitoring.py": "442c2a9c75e8c00b30a8f2e91cb52d87c466b19a80aea9128b10c021e61b922e",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase3_item1b.py": "b4d78f4384651711fe83ff00eb6719c6c61664699d801ecb34978b9546439da9",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_h2_common.py": "79918e7de887d1f0a2de5523156b27d00dc2d85ac930a07efc34d817d904fe65",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_h2_prepare.py": "f657be036dc48fc5936f56c4b8510a4b295d4787af6877cc5ccaad663107be17",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_recovery_monitor.py": "4c423f8b0dff69304d0046c0a9c319bb31aa0c6d1cbaeb3942726b5e6caa0496",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_recovery_supervisor.py": "5ccb2a6eec95b0700b2063a2f8df414037528e90c206648c1c5e49c5e89e741d",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_common.py": "38a29a34663a5d6310571eeff53b5d18a0355e51b9fa58a08033f64fa379a9df",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_common.py": "4e85f5d211c4d3e2188422aa4c0cefa40dcfa14ef32e90894ed3412de1fd1d12",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_data.py": "fe3aa3e023f22f4fe55cd47e7ffa677b58f52a041867ac05620cbd329d913cb6",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_guard.py": "7144fa77ffe61a2a3c92c050234e370ef4dfeb0e31fafb419f752f969737ff1a",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_loss.py": "79e291866bda300ecdc5577c62903d34abeec14f5cc36db05e346f9ffd64c24b",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_queue.py": "a04879cc0169ecb523a88360cddca1e234d91b3adbc20b3ec31d02461ed767eb",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_worker.py": "0f8f35e1422021b2a4f097c2e62267fb935fe4658fd0ee8ccd251609a4062ea7",
|
||||
"/home/frank14f/optane/MeltMind/tools/prepare_phase3_item_c.py": "782c415b1f536eb2b85aa2c9c4a9a0dc278f9b7134e8a83e2a6af8199396c53a",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_custody_common.py": "588c646101be0d50f505f9a12a1c98361aa3437c8c594eecb7321831f717a741",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_custody_statistics.py": "188565801f6901d0192f79aec0889e632501e672a4149d76d1835efbb9b45447",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase3_item_c.py": "3115aa9049974eac86657aa4cee7b5d716f6ee93268555cc86242029ebefce4c",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase2_h2_s2.py": "4235890fd194b7778e937f0be9245b964512df2c75d962ef703e3a988c5b8805",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_guard.py": "6bc7624b0cfe850591f77153cd2034eefdd62f7e63b93211ccdb87eb8aba346c",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_common.py": "308486507ae3589a0e6cc40565f6e60e6d2f63ae74c4d828bdacf33b3b8b8062",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_fixed_io.py": "ea17fdfc7508f3f980e3cdf69534b1eb8ceb8a8e469c32451b2d5e16d0ada575",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_worker.py": "37dfd75fff50b3d65f4ef37f4255536616582fe5f3d01155a0e037325ab5c7cd",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_worker.py": "ecee04b183ef2b93b7b39a447d48c40a5843d3190e48d2d7eb4939811a5db9b0",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_prepare.py": "964fae13ac95e4de4dc50b3c0a80791d86349ce3a9f3c5a6e7dc3cf8ca041fad",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_prepare.py": "3a31af0766f628a80f38e0b7286beb11ca3109188eabc4adfdb2ecb0eef3dc27",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_queue.py": "d82dfb365956e08250758bb4c651a19fbecce92fe209072f5f8ed35970450735",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_launch.py": "480fac25dea5a2006bb84130a4b0a0b253b3040b2150a56c38eada62b2c259e6",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_guard.py": "ae94711adc17a7630f398e8d4120fa3cc8f60979a00122c4cd2dc364c382ced5",
|
||||
"/home/frank14f/optane/MeltMind/docs/phase2-h2-s2-execution-contract-2026-11-17.md": "1e8e5ef31650519737cd6cd247abc0bd04744dd93ffb013eff88866d42f03073"
|
||||
},
|
||||
"verdict": "PASS",
|
||||
"reviewer": "MeltAssist",
|
||||
"scope": "Independent item1b five-fold100epoch training-entry review under Claude1795010666; reused fixed capacity rule, not c entry PASS",
|
||||
"evidence_sha256": "ea3c1fc443e312afeb037a82e2b5786443a454c1d2732716ef2e459869f39bd3",
|
||||
"audit_source_sha256": "0883429df7b4a294c3ddd0635f10b1f9ad073c024a97c8f6c9eff807f7f58609",
|
||||
"findings": [],
|
||||
"passed_checks": [
|
||||
"Exact60code/188source/74local immutable Git files; c52+seed1 51+TrackA27 pins all unchanged.",
|
||||
"Original55ffc6f9 schedule implemented exactly: seed0,5warmup epochs,95cosine decay epochs,1e-4/1e-3/1e-5 endpoints, only finalepoch100; no reuse of original200epoch midpoint.",
|
||||
"Original19inputs/H2/MSE/optimizer/weight decay/clipping/seed0/part-order/learning normalization retained. Training and validation loop AST equals original S2 after declared100epoch and per-fold cost-gate substitutions.",
|
||||
"Original five paired200epoch seed0 model/result/learning-normalization identities verified by metadata and opaque hashes; original historical diagnostic validation timing maxima equal new job bounds. No model/target arrays decoded.",
|
||||
"Per10epoch full-node validation MSE remains diagnostic only. Fold1 same-query-domain component-equal relativeL2 screen is separate and unchanged; delta<=-.001 and abs effect strictly greater than same-domain seed1-seed0, missingreference waits, no newseed.",
|
||||
"Five folds may train before screen per0666. Queue stops at TRAIN_COMPLETE_VALIDATION_SCREEN_PENDING with five frozen models; failed screen archives without test, and test postfit still needs actual separate pre-score review.",
|
||||
"Final test contract preserves552parts/204components,2000paired component bootstrap seed2026111408,95CI, upperdelta strictly<-.001 and abs pooled effect>abs matched seed difference; equality/incomplete fail, development reuse limitations retained.",
|
||||
"Shared adapter AST differs only item1b registration key/cap88560/GPU0..2; admission/registry/transfer/release/validate are same reviewed functions. Mixed c/1b pending reservation blocks overadmission; GPU0 oldpending, GPU3single and CPU/memory rules retained.",
|
||||
"88560 includes post7200GPU/3600CPU, prep600 and per-fold60setup/wait/failure accounting. Actual30000 runaway independent of projected fold cost; total1.5 remaining train+diagnostic cost,120cleanup; phase3<=720000 and TrackA separate.",
|
||||
"Failure/cleanup/late-peak accounting and model freeze inherited unchanged from reviewed c orchestration except declared1b paths/100epoch status. No retries.",
|
||||
"Independent23CPU synthetic tests PASS; measured full childCPU4.113682s; author pre-freeze missingGIB failure and corrected log preserved."
|
||||
],
|
||||
"limitations": [
|
||||
"This PASS binds only45b5df30/d33f036a and full60code map; it is not a scientific result or reuse of c entry PASS.",
|
||||
"No training launch, GPU query, real scientific array/label/model value decode, live source/registry/Git modification. Only own review artifacts and temporary synthetic fixtures written.",
|
||||
"Validation screen/prediction/scoring implementation is not included or approved here. Actual screen, seedreference matching, complete fivefold termination and finalcost gates must hold before separate reviewed test postfit.",
|
||||
"The original corpus has development exposure; one seed difference does not estimate all training randomness.",
|
||||
"Actual shared-card training timings/memory remain runtime evidence; the1.5factor and engineering peaks are limits/projections rather than new1b measurements.",
|
||||
"Publicationffd5c929 differs from implementation45b5df30. Local Git bytes verified; no independent remote raw-byte verification claim."
|
||||
]
|
||||
}
|
||||
@@ -0,0 +1,28 @@
|
||||
test_actual_runaway_stops (tools.test_phase3_item1b.BudgetTests) ... ok
|
||||
test_incomplete_history_rejected (tools.test_phase3_item1b.BudgetTests) ... ok
|
||||
test_initial_budget_includes_postfit (tools.test_phase3_item1b.BudgetTests) ... ok
|
||||
test_late_validation_max_used (tools.test_phase3_item1b.BudgetTests) ... ok
|
||||
test_no_single_fold_projection_gate (tools.test_phase3_item1b.BudgetTests) ... ok
|
||||
test_remaining_validation_charged (tools.test_phase3_item1b.BudgetTests) ... ok
|
||||
test_failure_no_retry_or_test (tools.test_phase3_item1b.QueueTests) ... ok
|
||||
test_five_fits_before_screen_but_no_test (tools.test_phase3_item1b.QueueTests) ... ok
|
||||
test_complete_warmup_and_decay (tools.test_phase3_item1b.ScheduleTests) ... ok
|
||||
test_exact_endpoints (tools.test_phase3_item1b.ScheduleTests) ... ok
|
||||
test_incomplete_test_not_confirmation (tools.test_phase3_item1b.ScheduleTests) ... ok
|
||||
test_invalid_epoch (tools.test_phase3_item1b.ScheduleTests) ... ok
|
||||
test_no_seed_no_decision (tools.test_phase3_item1b.ScheduleTests) ... ok
|
||||
test_nonfinite_comparisons_rejected (tools.test_phase3_item1b.ScheduleTests) ... ok
|
||||
test_not_original_midpoint (tools.test_phase3_item1b.ScheduleTests) ... ok
|
||||
test_only_fixed_last_checkpoint (tools.test_phase3_item1b.ScheduleTests) ... ok
|
||||
test_screen_exact_boundary (tools.test_phase3_item1b.ScheduleTests) ... ok
|
||||
test_test_strict_boundary_and_seed (tools.test_phase3_item1b.ScheduleTests) ... ok
|
||||
test_worker_original_train_step_and_seed0 (tools.test_phase3_item1b.ScheduleTests) ... ok
|
||||
test_mixed_pending_c_reservation_blocks_1b (tools.test_phase3_item1b.SharedWrapperTests) ... ok
|
||||
test_mixed_two_visible_allows_only_one_new (tools.test_phase3_item1b.SharedWrapperTests) ... ok
|
||||
test_shared_c_and_1b_registration_independent (tools.test_phase3_item1b.SharedWrapperTests) ... ok
|
||||
test_wrapper_changes_only_key_cap_gpu_set (tools.test_phase3_item1b.SharedWrapperTests) ... ok
|
||||
|
||||
----------------------------------------------------------------------
|
||||
Ran 23 tests in 3.182s
|
||||
|
||||
OK
|
||||
@@ -0,0 +1,10 @@
|
||||
{
|
||||
"status": "PASS",
|
||||
"returncode": 0,
|
||||
"wall_seconds": 5.96043398801703,
|
||||
"cpu_seconds": 5.9348849999999995,
|
||||
"synthetic_only": true,
|
||||
"real_arrays_decoded": false,
|
||||
"models_loaded": false,
|
||||
"gpu_query_or_workload": false
|
||||
}
|
||||
@@ -0,0 +1,11 @@
|
||||
{
|
||||
"status": "PASS",
|
||||
"returncode": 0,
|
||||
"wall_seconds": 5.801306594046764,
|
||||
"cpu_seconds": 5.759417,
|
||||
"synthetic_only": true,
|
||||
"real_arrays_decoded": false,
|
||||
"models_loaded": false,
|
||||
"gpu_query_or_workload": false,
|
||||
"change": "Author inspection caught copied guard final epoch100 mismatch; corrected to200 before freeze and added assertion; prior28 tests retained"
|
||||
}
|
||||
@@ -0,0 +1,33 @@
|
||||
test_actual_runaway_stops (test_phase3_item_b.BudgetTests) ... ok
|
||||
test_incomplete_history_rejected (test_phase3_item_b.BudgetTests) ... ok
|
||||
test_initial_budget_includes_postfit (test_phase3_item_b.BudgetTests) ... ok
|
||||
test_late_validation_max_used (test_phase3_item_b.BudgetTests) ... ok
|
||||
test_no_single_fold_projection_gate (test_phase3_item_b.BudgetTests) ... ok
|
||||
test_remaining_validation_charged (test_phase3_item_b.BudgetTests) ... ok
|
||||
test_all_learning_invalid_stops (test_phase3_item_b.ContextTests) ... ok
|
||||
test_infinity_or_wrong_shape_stops (test_phase3_item_b.ContextTests) ... ok
|
||||
test_joint_missing_is_zero_after_normalization (test_phase3_item_b.ContextTests) ... ok
|
||||
test_learning_only_iteration (test_phase3_item_b.ContextTests) ... ok
|
||||
test_original_role_guard_precedes_context_read (test_phase3_item_b.ContextTests) ... ok
|
||||
test_partial_missing_stops (test_phase3_item_b.ContextTests) ... ok
|
||||
test_status_mismatch_not_scientific_missing (test_phase3_item_b.ContextTests) ... ok
|
||||
test_unit_node_moments_not_equal_part_means (test_phase3_item_b.ContextTests) ... ok
|
||||
test_valid_mean_still_has_flag (test_phase3_item_b.ContextTests) ... ok
|
||||
test_zero_variance_scale_one (test_phase3_item_b.ContextTests) ... ok
|
||||
test_all_shared_parameters_and_rng_match (test_phase3_item_b.ModelTests) ... ok
|
||||
test_exactly384_additional_parameters (test_phase3_item_b.ModelTests) ... ok
|
||||
test_new_columns_can_learn_without_changing_recipe (test_phase3_item_b.ModelTests) ... ok
|
||||
test_nonzero_readout_still_matches_original_initial_function (test_phase3_item_b.ModelTests) ... ok
|
||||
test_worker_uses_original200_recipe (test_phase3_item_b.ModelTests) ... ok
|
||||
test_wrong_input_columns_rejected (test_phase3_item_b.ModelTests) ... ok
|
||||
test_failure_no_retry_or_test (test_phase3_item_b.QueueTests) ... ok
|
||||
test_five_fits_before_screen_but_no_test (test_phase3_item_b.QueueTests) ... ok
|
||||
test_mixed_pending_c_reservation_blocks_1b (test_phase3_item_b.SharedWrapperTests) ... ok
|
||||
test_mixed_two_visible_allows_only_one_new (test_phase3_item_b.SharedWrapperTests) ... ok
|
||||
test_shared_c_and_1b_registration_independent (test_phase3_item_b.SharedWrapperTests) ... ok
|
||||
test_wrapper_changes_only_key_cap_gpu_set (test_phase3_item_b.SharedWrapperTests) ... ok
|
||||
|
||||
----------------------------------------------------------------------
|
||||
Ran 28 tests in 1.395s
|
||||
|
||||
OK
|
||||
@@ -0,0 +1,34 @@
|
||||
test_actual_runaway_stops (test_phase3_item_b.BudgetTests) ... ok
|
||||
test_incomplete_history_rejected (test_phase3_item_b.BudgetTests) ... ok
|
||||
test_initial_budget_includes_postfit (test_phase3_item_b.BudgetTests) ... ok
|
||||
test_late_validation_max_used (test_phase3_item_b.BudgetTests) ... ok
|
||||
test_no_single_fold_projection_gate (test_phase3_item_b.BudgetTests) ... ok
|
||||
test_remaining_validation_charged (test_phase3_item_b.BudgetTests) ... ok
|
||||
test_all_learning_invalid_stops (test_phase3_item_b.ContextTests) ... ok
|
||||
test_infinity_or_wrong_shape_stops (test_phase3_item_b.ContextTests) ... ok
|
||||
test_joint_missing_is_zero_after_normalization (test_phase3_item_b.ContextTests) ... ok
|
||||
test_learning_only_iteration (test_phase3_item_b.ContextTests) ... ok
|
||||
test_original_role_guard_precedes_context_read (test_phase3_item_b.ContextTests) ... ok
|
||||
test_partial_missing_stops (test_phase3_item_b.ContextTests) ... ok
|
||||
test_status_mismatch_not_scientific_missing (test_phase3_item_b.ContextTests) ... ok
|
||||
test_unit_node_moments_not_equal_part_means (test_phase3_item_b.ContextTests) ... ok
|
||||
test_valid_mean_still_has_flag (test_phase3_item_b.ContextTests) ... ok
|
||||
test_zero_variance_scale_one (test_phase3_item_b.ContextTests) ... ok
|
||||
test_all_shared_parameters_and_rng_match (test_phase3_item_b.ModelTests) ... ok
|
||||
test_exactly384_additional_parameters (test_phase3_item_b.ModelTests) ... ok
|
||||
test_guard_accepts_only_complete200_epoch_identity (test_phase3_item_b.ModelTests) ... ok
|
||||
test_new_columns_can_learn_without_changing_recipe (test_phase3_item_b.ModelTests) ... ok
|
||||
test_nonzero_readout_still_matches_original_initial_function (test_phase3_item_b.ModelTests) ... ok
|
||||
test_worker_uses_original200_recipe (test_phase3_item_b.ModelTests) ... ok
|
||||
test_wrong_input_columns_rejected (test_phase3_item_b.ModelTests) ... ok
|
||||
test_failure_no_retry_or_test (test_phase3_item_b.QueueTests) ... ok
|
||||
test_five_fits_stop_before_postfit (test_phase3_item_b.QueueTests) ... ok
|
||||
test_mixed_pending_c_reservation_blocks_b (test_phase3_item_b.SharedWrapperTests) ... ok
|
||||
test_mixed_two_visible_allows_only_one_new (test_phase3_item_b.SharedWrapperTests) ... ok
|
||||
test_shared_c_and_b_registration_independent (test_phase3_item_b.SharedWrapperTests) ... ok
|
||||
test_wrapper_changes_only_key_cap_gpu_set (test_phase3_item_b.SharedWrapperTests) ... ok
|
||||
|
||||
----------------------------------------------------------------------
|
||||
Ran 29 tests in 1.197s
|
||||
|
||||
OK
|
||||
@@ -0,0 +1,12 @@
|
||||
{
|
||||
"status": "PASS",
|
||||
"binding_sha256": "e5b01d81f8bfee991dee9a2971f8048d77b6c15ab61e3de08b2f97385f413261",
|
||||
"reviewed_commit": "b2a9dfdc4a2c237416452ff6b4c32bdbf34e172e",
|
||||
"contract_sha256": "df80f3b4bbb0d4ee285401ff50b40cfc6bfd06f3302abcd53b41f14a20a204de",
|
||||
"implementation_proof_sha256": "0501275106a0698c9dac6f75554b570bb82f7d4640e76b84f51cf759a8bdadd2",
|
||||
"comparison_contract_sha256": "06556bf1f7bd5a1a79451bd9eed47fc6ec027e7e06bc63c6f490a50dfbee6d20",
|
||||
"code_count": 69,
|
||||
"source_count": 1300,
|
||||
"started": false,
|
||||
"review_adopted": false
|
||||
}
|
||||
@@ -0,0 +1,28 @@
|
||||
{
|
||||
"status": "PASS",
|
||||
"existing_pins": [
|
||||
{
|
||||
"root": "/home/frank14f/MeltMind-runs/phase3-item-c-2026-11-18",
|
||||
"code_count": 52,
|
||||
"status": "PASS"
|
||||
},
|
||||
{
|
||||
"root": "/home/frank14f/MeltMind-runs/phase3-seed1-v2-2026-11-18",
|
||||
"code_count": 51,
|
||||
"status": "PASS"
|
||||
},
|
||||
{
|
||||
"root": "/home/frank14f/MeltMind-runs/track-a-recovery-2026-11-18",
|
||||
"code_count": 27,
|
||||
"status": "PASS"
|
||||
},
|
||||
{
|
||||
"root": "/home/frank14f/MeltMind-runs/phase3-item-1b-2026-11-18",
|
||||
"code_count": 60,
|
||||
"status": "PASS"
|
||||
}
|
||||
],
|
||||
"code_count": 69,
|
||||
"source_count": 1300,
|
||||
"unix": 1795012287.384036
|
||||
}
|
||||
@@ -0,0 +1,291 @@
|
||||
{
|
||||
"unix": 1795011943.081557,
|
||||
"snapshot": {
|
||||
"devices": {
|
||||
"GPU-e8853ae7-a302-6a17-bcfa-aa767e9ed20c": {
|
||||
"total": 17179869184,
|
||||
"used": 1928331264
|
||||
},
|
||||
"GPU-f1c9fd5a-e917-56e0-21fb-24035ba44a36": {
|
||||
"total": 17179869184,
|
||||
"used": 5390729216
|
||||
},
|
||||
"GPU-b8c11c01-c91f-aa0b-abf6-dad35085cc76": {
|
||||
"total": 17179869184,
|
||||
"used": 5770313728
|
||||
},
|
||||
"GPU-8b71b032-1b75-2891-935c-ce106a53034a": {
|
||||
"total": 17179869184,
|
||||
"used": 1928331264
|
||||
}
|
||||
},
|
||||
"processes": [
|
||||
{
|
||||
"gpu_uuid": "GPU-e8853ae7-a302-6a17-bcfa-aa767e9ed20c",
|
||||
"pid": 1765688,
|
||||
"vram": 1918894080
|
||||
},
|
||||
{
|
||||
"gpu_uuid": "GPU-f1c9fd5a-e917-56e0-21fb-24035ba44a36",
|
||||
"pid": 1476726,
|
||||
"vram": 1793064960
|
||||
},
|
||||
{
|
||||
"gpu_uuid": "GPU-f1c9fd5a-e917-56e0-21fb-24035ba44a36",
|
||||
"pid": 1755848,
|
||||
"vram": 1793064960
|
||||
},
|
||||
{
|
||||
"gpu_uuid": "GPU-f1c9fd5a-e917-56e0-21fb-24035ba44a36",
|
||||
"pid": 1755878,
|
||||
"vram": 1795162112
|
||||
},
|
||||
{
|
||||
"gpu_uuid": "GPU-b8c11c01-c91f-aa0b-abf6-dad35085cc76",
|
||||
"pid": 1650207,
|
||||
"vram": 1918894080
|
||||
},
|
||||
{
|
||||
"gpu_uuid": "GPU-b8c11c01-c91f-aa0b-abf6-dad35085cc76",
|
||||
"pid": 1756022,
|
||||
"vram": 1920991232
|
||||
},
|
||||
{
|
||||
"gpu_uuid": "GPU-b8c11c01-c91f-aa0b-abf6-dad35085cc76",
|
||||
"pid": 1755953,
|
||||
"vram": 1920991232
|
||||
},
|
||||
{
|
||||
"gpu_uuid": "GPU-8b71b032-1b75-2891-935c-ce106a53034a",
|
||||
"pid": 1598144,
|
||||
"vram": 1918894080
|
||||
}
|
||||
],
|
||||
"ram_total": 67378212864,
|
||||
"ram_available": 51724812288,
|
||||
"cpu_percent": 25.5,
|
||||
"old_pending": true,
|
||||
"unix": 1795011942.9723783
|
||||
},
|
||||
"leases": [
|
||||
{
|
||||
"token": "da7989f6fb454ec99e83ffa8092b056a",
|
||||
"state": "GUARD_BOUND",
|
||||
"root": "/home/frank14f/MeltMind-runs/phase3-item-c-2026-11-18",
|
||||
"fold": 0,
|
||||
"contract_sha256": "0e8533194b9987d234a8ea419a2d4e1a3993ab18c41400d5a89d0beaacf6a465",
|
||||
"gpu_uuid": "GPU-f1c9fd5a-e917-56e0-21fb-24035ba44a36",
|
||||
"worker_cpus": [
|
||||
16
|
||||
],
|
||||
"owner": {
|
||||
"pid": 1755765,
|
||||
"created": 1795010605.64
|
||||
},
|
||||
"queue_owner": {
|
||||
"pid": 1755713,
|
||||
"created": 1795010602.48
|
||||
},
|
||||
"vram_reserve": 4294967296,
|
||||
"rss_reserve": 4294967296,
|
||||
"created_unix": 1795010605.6536286,
|
||||
"observed_pids": [
|
||||
1755765,
|
||||
1755848
|
||||
],
|
||||
"observed_rss": 1805930496
|
||||
},
|
||||
{
|
||||
"token": "a7e8a5931ff844faa5a61f420e155704",
|
||||
"state": "GUARD_BOUND",
|
||||
"root": "/home/frank14f/MeltMind-runs/phase3-item-c-2026-11-18",
|
||||
"fold": 1,
|
||||
"contract_sha256": "0e8533194b9987d234a8ea419a2d4e1a3993ab18c41400d5a89d0beaacf6a465",
|
||||
"gpu_uuid": "GPU-f1c9fd5a-e917-56e0-21fb-24035ba44a36",
|
||||
"worker_cpus": [
|
||||
17
|
||||
],
|
||||
"owner": {
|
||||
"pid": 1755824,
|
||||
"created": 1795010607.05
|
||||
},
|
||||
"queue_owner": {
|
||||
"pid": 1755713,
|
||||
"created": 1795010602.48
|
||||
},
|
||||
"vram_reserve": 4294967296,
|
||||
"rss_reserve": 4294967296,
|
||||
"created_unix": 1795010607.0678425,
|
||||
"observed_pids": [
|
||||
1755824,
|
||||
1755878
|
||||
],
|
||||
"observed_rss": 1775656960
|
||||
},
|
||||
{
|
||||
"token": "ded7a54470cb409b85604b8d7248abc3",
|
||||
"state": "GUARD_BOUND",
|
||||
"root": "/home/frank14f/MeltMind-runs/phase3-item-c-2026-11-18",
|
||||
"fold": 2,
|
||||
"contract_sha256": "0e8533194b9987d234a8ea419a2d4e1a3993ab18c41400d5a89d0beaacf6a465",
|
||||
"gpu_uuid": "GPU-b8c11c01-c91f-aa0b-abf6-dad35085cc76",
|
||||
"worker_cpus": [
|
||||
18
|
||||
],
|
||||
"owner": {
|
||||
"pid": 1755862,
|
||||
"created": 1795010608.29
|
||||
},
|
||||
"queue_owner": {
|
||||
"pid": 1755713,
|
||||
"created": 1795010602.48
|
||||
},
|
||||
"vram_reserve": 4294967296,
|
||||
"rss_reserve": 4294967296,
|
||||
"created_unix": 1795010608.3074121,
|
||||
"observed_pids": [
|
||||
1755862,
|
||||
1755953
|
||||
],
|
||||
"observed_rss": 1790693376
|
||||
},
|
||||
{
|
||||
"token": "53f07e4a34204fd8a6641002c8d29f3e",
|
||||
"state": "GUARD_BOUND",
|
||||
"root": "/home/frank14f/MeltMind-runs/phase3-item-c-2026-11-18",
|
||||
"fold": 3,
|
||||
"contract_sha256": "0e8533194b9987d234a8ea419a2d4e1a3993ab18c41400d5a89d0beaacf6a465",
|
||||
"gpu_uuid": "GPU-b8c11c01-c91f-aa0b-abf6-dad35085cc76",
|
||||
"worker_cpus": [
|
||||
19
|
||||
],
|
||||
"owner": {
|
||||
"pid": 1755893,
|
||||
"created": 1795010609.69
|
||||
},
|
||||
"queue_owner": {
|
||||
"pid": 1755713,
|
||||
"created": 1795010602.48
|
||||
},
|
||||
"vram_reserve": 4294967296,
|
||||
"rss_reserve": 4294967296,
|
||||
"created_unix": 1795010609.7058253,
|
||||
"observed_pids": [
|
||||
1755893,
|
||||
1756022
|
||||
],
|
||||
"observed_rss": 1809506304
|
||||
}
|
||||
],
|
||||
"admission": {
|
||||
"GPU-e8853ae7-a302-6a17-bcfa-aa767e9ed20c": [
|
||||
false,
|
||||
"GPU0_RESERVED_FOR_OLD_PENDING"
|
||||
],
|
||||
"GPU-f1c9fd5a-e917-56e0-21fb-24035ba44a36": [
|
||||
false,
|
||||
"GPU_SLOT_LIMIT"
|
||||
],
|
||||
"GPU-b8c11c01-c91f-aa0b-abf6-dad35085cc76": [
|
||||
false,
|
||||
"GPU_SLOT_LIMIT"
|
||||
]
|
||||
},
|
||||
"free_worker_cpus": [
|
||||
20,
|
||||
21,
|
||||
22,
|
||||
23,
|
||||
24,
|
||||
25
|
||||
],
|
||||
"old_pins_and_progress": [
|
||||
{
|
||||
"root": "/home/frank14f/MeltMind-runs/phase3-item-c-2026-11-18",
|
||||
"code_count": 52,
|
||||
"progress": {
|
||||
"unix": 1795011941.910367,
|
||||
"stage": "ITEM_C_TRAINING",
|
||||
"active": [
|
||||
0,
|
||||
1,
|
||||
2,
|
||||
3
|
||||
],
|
||||
"pending": [
|
||||
4
|
||||
],
|
||||
"terminal": [],
|
||||
"charged_training_gpu_seconds": 7582.2292879448505,
|
||||
"postfit_started": false
|
||||
},
|
||||
"terminal": false
|
||||
},
|
||||
{
|
||||
"root": "/home/frank14f/MeltMind-runs/phase3-seed1-v2-2026-11-18",
|
||||
"code_count": 51,
|
||||
"progress": {
|
||||
"stage": "SEED1_TRAINING",
|
||||
"unix": 1795011942.2254956,
|
||||
"active": [
|
||||
{
|
||||
"id": 1,
|
||||
"guard_pid": 1476628
|
||||
},
|
||||
{
|
||||
"id": 2,
|
||||
"guard_pid": 1598086
|
||||
},
|
||||
{
|
||||
"id": 3,
|
||||
"guard_pid": 1765488
|
||||
}
|
||||
],
|
||||
"terminal": [
|
||||
{
|
||||
"id": 0,
|
||||
"status": "PASS",
|
||||
"returncode": 0,
|
||||
"gpu_seconds_charged": 18897.35776199703,
|
||||
"guard_complete_sha256": "74043e45ac73a511be69bfe0d51c170752a7c8aecf4a46e12423c89baa174d70",
|
||||
"gpu_uuid": "GPU-e8853ae7-a302-6a17-bcfa-aa767e9ed20c",
|
||||
"assignment_sha256": "498011b74b369fb6c4144659d9ea7449863205446329dc36b08e60a78e0d8134"
|
||||
}
|
||||
],
|
||||
"pending": [
|
||||
4
|
||||
],
|
||||
"charged_training_gpu_seconds": 51869.11572699901,
|
||||
"postfit_gpu_reserve": 7200
|
||||
},
|
||||
"terminal": false
|
||||
},
|
||||
{
|
||||
"root": "/home/frank14f/MeltMind-runs/track-a-recovery-2026-11-18",
|
||||
"code_count": 27,
|
||||
"progress": {
|
||||
"unix": 1795011942.1222477,
|
||||
"active": [
|
||||
{
|
||||
"id": 4,
|
||||
"guard_pid": 1650109
|
||||
}
|
||||
],
|
||||
"pending": [
|
||||
3
|
||||
],
|
||||
"terminal": [],
|
||||
"old_gpu_seconds_once": 47943.94719455298,
|
||||
"new_training_gpu_seconds": 15592.43198554893,
|
||||
"cumulative_training_gpu_seconds": 63536.37919126195,
|
||||
"postfit_started": false
|
||||
},
|
||||
"terminal": false
|
||||
}
|
||||
],
|
||||
"duplicates": [],
|
||||
"unit": "MainPID=0\nLoadState=not-found\nActiveState=inactive\n",
|
||||
"entry_registered": false,
|
||||
"started": false,
|
||||
"free_disk": 237145300992
|
||||
}
|
||||
@@ -0,0 +1,10 @@
|
||||
{
|
||||
"status": "PASS",
|
||||
"review_sha256": "4ea5b84df28d7dc2c72318bf315f9220cec45b27cfe46bfb058b8f9719d1a004",
|
||||
"binding_sha256": "13a5142ee9d1d635b9d7f2979c2cd392de8feda8fcca722fcdc469a2d9927873",
|
||||
"review_freeze": "ec17fde0384d30e44324c1d85eaf794ac33ee0a5",
|
||||
"claude_acceptance": "1795011859.371609",
|
||||
"native_delivery_received": true,
|
||||
"started": false,
|
||||
"unix": 1795011918.4847505
|
||||
}
|
||||
@@ -789,3 +789,17 @@ implementation 45b5df30098f74dc3c93a2f54365d7d866b810e0,contract d33f036adc48c
|
||||
|
||||
c原队列继续,最新读取1795011458:4active/1pending,无terminal,投影163798.657995/205200秒PASS。后续以真实回执为准,不重复启动或修改pins。用户直接计算授权已归档,无需再问。审阅请求须TRON原生送达Assist并收到其确认;Slack只作Claude/公共记录,不冒称已送达或已唤醒其他Astra。
|
||||
|
||||
|
||||
|
||||
## 最新 1b 已采用实际PASS待容量;b 入口已固定待审(2026-11-18)
|
||||
|
||||
TRON原生转交Assist1795011841已实收,四原件SHA核对,1b原件review4ea5b84d完整60code绑定已采用,review freeze ec17fde0384d30e44324c1d85eaf794ac33ee0a5,validate_review/config PASS。Claude1795011859接受,原用户合并计算授权覆盖。实时审计17950119xx:GPU0/1/2/3为1/3/3/1,GPU1/2 SLOT_LIMIT,GPU0为旧pending保留;无1b登记/started/进程,unit not-found。不是启动失败,也不是科学失败。
|
||||
|
||||
Claude1795011981及TRON后续原生补充均已实读:不提前启动1b队列等待计费;先由seed1/TrackA/c现有队列自己的派发占用待运行槽,1b在GPU1/2后续真正空槽一次启动并审计,不抢容量/不改共享规则。1b只训五折,到TRAIN_COMPLETE_VALIDATION_SCREEN_PENDING;筛选与test后处理保持原门。见artifacts/phase3-item1b-launch-2026-11-18。
|
||||
|
||||
依Claude1607/1859继续的b入口已完成:Mold规格de120e9d全文归档;原H2只增两列B13均值与有效位,共384零初始权重,旧参数/RNG保持;seed0/200轮/全552名单不变。既有缓存元数据510有效/42科学缺失,仅读哈希,无新分区/真实数组解码。29项合成PASS,作者冻结前发现并修正复制guard的100轮条件、补200轮断言,原两次验收日志保留。
|
||||
|
||||
b implementation b2a9dfdc4a2c237416452ff6b4c32bdbf34e172e,contract df80f3b4bbb0d4ee285401ff50b40cfc6bfd06f3302abcd53b41f14a20a204de,proof 0501275106a0698c9dac6f75554b570bb82f7d4640e76b84f51cf759a8bdadd2,comparison 06556bf1f7bd5a1a79451bd9eed47fc6ec027e7e06bc63c6f490a50dfbee6d20,binding e5b01d81f8bfee991dee9a2971f8048d77b6c15ab61e3de08b2f97385f413261,69code/1300source。唯一根/home/frank14f/MeltMind-runs/phase3-item-b-2026-11-18,报告docs/phase3-item-b-entry-2026-11-18.md。178200GPU秒含post7200/CPU3600,phase3累计720000;共享容量逻辑保持原已审版。config/五字段本地核验PASS不代表Assist PASS,无review采用/started/训练/预测/评分。
|
||||
|
||||
原c/seed1/TrackA及固定1b共190代码pins核一致;不重启旧队列。后续b须TRON原生送达Assist并确认收件后取得实际入口审阅,不能把Slack公共记录当送达。当前1b待容量、b待审,不重问覆盖授权。
|
||||
|
||||
@@ -1,3 +1,17 @@
|
||||
## 最新 1b 已采用实际PASS待容量;b 入口已固定待审(2026-11-18)
|
||||
|
||||
TRON原生转交Assist1795011841已实收,四原件SHA核对,1b原件review4ea5b84d完整60code绑定已采用,review freeze ec17fde0384d30e44324c1d85eaf794ac33ee0a5,validate_review/config PASS。Claude1795011859接受,原用户合并计算授权覆盖。实时审计17950119xx:GPU0/1/2/3为1/3/3/1,GPU1/2 SLOT_LIMIT,GPU0为旧pending保留;无1b登记/started/进程,unit not-found。不是启动失败,也不是科学失败。
|
||||
|
||||
Claude1795011981及TRON后续原生补充均已实读:不提前启动1b队列等待计费;先由seed1/TrackA/c现有队列自己的派发占用待运行槽,1b在GPU1/2后续真正空槽一次启动并审计,不抢容量/不改共享规则。1b只训五折,到TRAIN_COMPLETE_VALIDATION_SCREEN_PENDING;筛选与test后处理保持原门。见artifacts/phase3-item1b-launch-2026-11-18。
|
||||
|
||||
依Claude1607/1859继续的b入口已完成:Mold规格de120e9d全文归档;原H2只增两列B13均值与有效位,共384零初始权重,旧参数/RNG保持;seed0/200轮/全552名单不变。既有缓存元数据510有效/42科学缺失,仅读哈希,无新分区/真实数组解码。29项合成PASS,作者冻结前发现并修正复制guard的100轮条件、补200轮断言,原两次验收日志保留。
|
||||
|
||||
b implementation b2a9dfdc4a2c237416452ff6b4c32bdbf34e172e,contract df80f3b4bbb0d4ee285401ff50b40cfc6bfd06f3302abcd53b41f14a20a204de,proof 0501275106a0698c9dac6f75554b570bb82f7d4640e76b84f51cf759a8bdadd2,comparison 06556bf1f7bd5a1a79451bd9eed47fc6ec027e7e06bc63c6f490a50dfbee6d20,binding e5b01d81f8bfee991dee9a2971f8048d77b6c15ab61e3de08b2f97385f413261,69code/1300source。唯一根/home/frank14f/MeltMind-runs/phase3-item-b-2026-11-18,报告docs/phase3-item-b-entry-2026-11-18.md。178200GPU秒含post7200/CPU3600,phase3累计720000;共享容量逻辑保持原已审版。config/五字段本地核验PASS不代表Assist PASS,无review采用/started/训练/预测/评分。
|
||||
|
||||
原c/seed1/TrackA及固定1b共190代码pins核一致;不重启旧队列。后续b须TRON原生送达Assist并确认收件后取得实际入口审阅,不能把Slack公共记录当送达。当前1b待容量、b待审,不重问覆盖授权。
|
||||
|
||||
---
|
||||
|
||||
## 最新 item 1b 入口已固定,待 Assist 实际审阅(2026-11-18)
|
||||
|
||||
Claude1795010666五折先训练/筛选仅作测试判决门已写入独立合同。100完整轮、seed0、原H2/19列/诊断保持;新88560GPU秒内含7200post预留,phase3累计720000与共享容量规则不变。仅新1b包装的登记键/帽/GPU0–2不同,原c/shared及seed1/TrackA共130代码pins均核PASS。
|
||||
|
||||
@@ -0,0 +1,15 @@
|
||||
# Phase 3 item b:H2 与通用八区背景
|
||||
|
||||
本入口回答原 H2 是否还能从两个 B13 区域均值及缺失标记获得增量。规格为 Mold `de120e9dc5527c01af5247a27c8cf529985c206821553bacb01be6071f2cdab5`,Claude1795006714允许各研究任务并行、seed1只约束最终判决;1795007580固定178200 GPU秒及phase3累计720000秒;1795011607/1859要求继续构建本入口。用户在本会话已直接批准所述执行范围,各入口仍须实际审阅通过。
|
||||
|
||||
唯一根 `/home/frank14f/MeltMind-runs/phase3-item-b-2026-11-18`。五折seed0/100%学习池/原200完整轮,原MSE、AdamW、衰减、裁剪、洗牌、FP64 cF跳接、最终检查点及每10轮validation诊断不变。原19列追加两列背景和共同有效0/1位,编码首层19→22、解码首层83→86,只增384权重。先完整建立原H2,再追加零权重,不额外抽随机数;全部共享初始参数与seed0对齐。初始函数等价不意味着训练后没有差异。
|
||||
|
||||
全部552件/204组件保留,复用B13-E1b固定缓存,不重新分区、不变K或种子。元数据及不透明哈希核验得到既有510件GEOMETRY_COMPLETE和42件SCIENTIFIC_MISSING;这是历史状态核对,没有新几何或数值验收。原结果/extra哈希、节点数、raw ID/原特征缓存、制备几何与查询名单身份逐件匹配。运行时仅允许完整两列有限或原科学缺失件的两列全NaN;单列缺失、局部缺失、inf、身份损坏均停止,不伪装成科学缺失。共同有效位保留在新输入中,收益是背景与缺失模式整体增量。
|
||||
|
||||
新列均值/总体标准差只由本折learning的有效监督节点按原H2单位节点权求得,零方差尺度1;无有效learning节点停止。无效件标准化后两列置0、标记0;有效值即使恰等于均值也标记1。原19列标准化路径不变。真实数值输入只在后续审阅通过的训练入口读取,测试标签没有入口路径。
|
||||
|
||||
训练队列复用已审共享容量规则,适配器仅登记键b/帽178200/GPU0–2不同;GPU0旧队列保留及GPU3单任务不变。总178200秒包含准备保守600、每折setup60、所有队列内等待/失败及postfit预留7200 GPU秒/3600 CPU秒。实际单折60000秒失控帽,总投影计剩余200轮训练及20次诊断,取实测最大时耗/历史验证时耗上界并乘1.5,另加cleanup;不设单折投影帽,首轮门及持续总帽超限停止,不重试。运行中的c、seed1、TrackA和已固定1b不改动。真实容量/实际显存及共享时耗仍须运行回执证明。
|
||||
|
||||
队列止于五个模型冻结和TRAIN_COMPLETE_POSTFIT_PENDING;不是评分入口。五折全部终态、原worker退出、最终冻结尾段账合格后,另需postfit事前审阅。判决保持552/204、2000配对组件bootstrap、seed2026111408、双侧95%;新减旧relative L2区间上界严格小于−0.001,且点效应绝对值严格超过同域seed1−seed0差,参照缺失等待。G有效/无效件和组件、分组差值、逐折、末点/Rg、tau、负时间及端到端成本完整报告,不能用子集替代全体。零缓存部署分区时延本入口未测,不能当零成本;同库开发暴露和单次seed差限制保留,不裁决独立类型/O1/真实连通假说。
|
||||
|
||||
29项CPU合成验收PASS:共享参数/RNG逐项对齐、384参数、初始函数等价、新列可学习、learning-only单位节点矩、零方差/共同缺失/标记与错误拒绝、原配方与200轮终态、含validation成本总帽、混合预留准入及五模型止步/失败零重试。初始28项测试未覆盖复制来的guard轮数条件,作者检查发现100/200不一致后在冻结前修正并补断言;两次原日志均保留。没有真实数组/标签/模型权重解码或GPU查询/计算。固定五字段绑定供MeltAssist一次独立入口审阅,未采用PASS、未启动。
|
||||
@@ -0,0 +1,4 @@
|
||||
{
|
||||
"contract_sha256": "d33f036adc48cde904fbb6bb0f615a5464f94b039906d0b45a4dbd15f0681bd3",
|
||||
"proof_sha256": "acb1740d76c6cb7a305fbd7c97e6b1f9e90b366fb5bccefe284e4ec58c183522"
|
||||
}
|
||||
@@ -0,0 +1,23 @@
|
||||
{
|
||||
"commit": "ec17fde0384d30e44324c1d85eaf794ac33ee0a5",
|
||||
"files": {
|
||||
"/home/frank14f/MeltMind-runs/phase3-item-1b-2026-11-18/contract.json": "d33f036adc48cde904fbb6bb0f615a5464f94b039906d0b45a4dbd15f0681bd3",
|
||||
"/home/frank14f/MeltMind-runs/phase3-item-1b-2026-11-18/expected-review-binding.json": "13a5142ee9d1d635b9d7f2979c2cd392de8feda8fcca722fcdc469a2d9927873",
|
||||
"/home/frank14f/MeltMind-runs/phase3-item-1b-2026-11-18/review.json": "4ea5b84df28d7dc2c72318bf315f9220cec45b27cfe46bfb058b8f9719d1a004",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/meltassist-item1b-entry-review-2026-11-18/review.json": "4ea5b84df28d7dc2c72318bf315f9220cec45b27cfe46bfb058b8f9719d1a004",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/meltassist-item1b-entry-review-2026-11-18/evidence.json": "ea3c1fc443e312afeb037a82e2b5786443a454c1d2732716ef2e459869f39bd3",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/meltassist-item1b-entry-review-2026-11-18/audit.py": "0883429df7b4a294c3ddd0635f10b1f9ad073c024a97c8f6c9eff807f7f58609",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/meltassist-item1b-entry-review-2026-11-18/synthetic-tests.log": "ff7715d22c8c0594fe184ae8da0a52ac5798f6651138d4ebe93493745bc385ab"
|
||||
},
|
||||
"git_paths": {
|
||||
"/home/frank14f/MeltMind-runs/phase3-item-1b-2026-11-18/contract.json": "home-runs/phase3-item-1b-2026-11-18/contract.json",
|
||||
"/home/frank14f/MeltMind-runs/phase3-item-1b-2026-11-18/expected-review-binding.json": "home-runs/phase3-item-1b-2026-11-18/expected-review-binding.json",
|
||||
"/home/frank14f/MeltMind-runs/phase3-item-1b-2026-11-18/review.json": "home-runs/phase3-item-1b-2026-11-18/review.json",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/meltassist-item1b-entry-review-2026-11-18/review.json": "artifacts/meltassist-item1b-entry-review-2026-11-18/review.json",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/meltassist-item1b-entry-review-2026-11-18/evidence.json": "artifacts/meltassist-item1b-entry-review-2026-11-18/evidence.json",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/meltassist-item1b-entry-review-2026-11-18/audit.py": "artifacts/meltassist-item1b-entry-review-2026-11-18/audit.py",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/meltassist-item1b-entry-review-2026-11-18/synthetic-tests.log": "artifacts/meltassist-item1b-entry-review-2026-11-18/synthetic-tests.log"
|
||||
},
|
||||
"remote_published": false,
|
||||
"local_repository": "/tmp/meltmind-phase2-local-fixed-20261117"
|
||||
}
|
||||
@@ -0,0 +1,95 @@
|
||||
{
|
||||
"reviewed_commit": "45b5df30098f74dc3c93a2f54365d7d866b810e0",
|
||||
"contract_sha256": "d33f036adc48cde904fbb6bb0f615a5464f94b039906d0b45a4dbd15f0681bd3",
|
||||
"implementation_proof_sha256": "73f28ef6d9991ef36f9e399b39ff24d849a656ee3caa8b3230e3c61732839f29",
|
||||
"comparison_contract_sha256": "648a99549689ae6aba02085d59fc5b9a10d350e48832c383570569a657d24aae",
|
||||
"code_sha256": {
|
||||
"/home/frank14f/optane/MeltMind/tools/freeze_g1_components.py": "d0cee60d95243ac2c59ef3f2244ba818315718c4e3d1895f2f9f064f1f7f5ecd",
|
||||
"/home/frank14f/optane/MeltMind/tools/gre24_models.py": "53b6fee8b700b9b982b06d29908ce2ce23e0c57512c46dbf74b2aade79e3ac3a",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_review.py": "915617aeb8292d03c654e637bc34a014c254fd178a773f135d1806c4b41372a8",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h1_worker_v2.py": "1543df85b615bb69b79bacd572074a7c77ae19dab110197601573a7e5805ed20",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_data.py": "7e03dd7a94e7bb43593277f183c222473d7cb7aaf2f60e53b45ab8d9499fcb09",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_model.py": "fab2470dd0654ca74c3b7d893f35ad1f3ae6c5c08ffe334ef6312cf6ec87892a",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_pilot_worker.py": "14651ca402e6d02466d8afdb77bc9e62fb2e752f0c0c3d20d98593157d5ebe81",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_home_freeze_io.py": "bf09204b85ed2a7038acdd202257f618474d1110a9d6d772f105a3ffb334d3ff",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_local_freeze_io.py": "60adf3382577613c5234f5cf30779336b9951b66e90c075cb7cd0adab9cc42dd",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_residual_math.py": "ffbd9624dec79bdaabf95244fe87209055908f31a977e73ea4377045d6f8fe62",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_capacity.py": "771faad0095712fd4fc8131ddcfabaa3901141592d2b055ea2faadf8ea3d3edf",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_common.py": "4e25effe8326ff8c7c3f751590315040209af054b32b0505dc3c6b001589db32",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_guard.py": "08b7d8ab2ac773fab8b749f9f61af259254422128e9634f8b715c81efce2651f",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_queue.py": "31dfd71369c63512c1d6fe176fd22e8db646520c92da533bb5530ca3843c388d",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_schedule.py": "06cb97594c3a87f42f712dcde6cec8c2d90e0b50a89a6be7954bdd2f5c18d5de",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_worker.py": "b781e56c519401bd38af6a522deb24506cba6993d2af008f85ea4e1715f1d9c6",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_seed1_v2_common.py": "25ca6be14027369393bd3616209b08f30e25892cccca83300483c5a116e8d75d",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_shared_capacity.py": "7c07b0aaf1bc4e6a980d016defcac326c3bd871134d2583bb4dcd88bb34728f8",
|
||||
"/home/frank14f/optane/MeltMind/tools/prepare_phase3_item1b.py": "c7bc9f3b7cf4137f52e1efad19dcb7045282d4e9a9c1aae50f066c9e488f7c01",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_gpu_idle_gate.py": "e2b1613e7e6ea398a8d555220a9365a1ad998ab4ef4b2bf8ec7cc4e31f8b4b9f",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_main_memory_guard.py": "e094f5443e11518efe9018215c11718a8e18aa462a977bf43e974b2d0b2aebb1",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/__init__.py": "a0fa1323d3553172da66c5d442c5fc4fcb0d5055c886c776928604550ea97609",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/attention.py": "7714adfccf6a0fc6b52a44aee3a50321a88898bf3d743680a1108c01e1ff8a61",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/blocks.py": "b13e986cea5f1d7f325a1638c1505197b4a105baacff5a9a515f23fcaf28ea75",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/features.py": "3b34ae61cf8012ebd58b7505a73d7aa43d8ee1890ac54f98d7917b67887a45a4",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/network.py": "5137e845d81d1223afc13bf87e03ccba1132a3baa75d5189a8f320b2752be071",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/normalization.py": "7180696d3845799e28d8c31ddef957d075de6a2e46d96a3a0f0df16fc05234a0",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/objectives.py": "5b6527a45f08690055d572335470ce5979e8bb86e1cab00de594f79ae21316b1",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_pilot_train.py": "e9cceead51a5cc092aad23409ad98909c84a62e7e10571e750d1e9be1ad3b948",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_precision_continuation_guard.py": "01235d897c7c03e5ffa109c58b14f5d24d128e8f1ba873e1f76cfb14e02f7a7f",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_recovery_gpu_supervisor.py": "689c0f3179e4fcd362e546d90214829a9e4f8739abdbdf74b560763e5a392f15",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_recovery_monitoring.py": "442c2a9c75e8c00b30a8f2e91cb52d87c466b19a80aea9128b10c021e61b922e",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase3_item1b.py": "b4d78f4384651711fe83ff00eb6719c6c61664699d801ecb34978b9546439da9",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_h2_common.py": "79918e7de887d1f0a2de5523156b27d00dc2d85ac930a07efc34d817d904fe65",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_h2_prepare.py": "f657be036dc48fc5936f56c4b8510a4b295d4787af6877cc5ccaad663107be17",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_recovery_monitor.py": "4c423f8b0dff69304d0046c0a9c319bb31aa0c6d1cbaeb3942726b5e6caa0496",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_recovery_supervisor.py": "5ccb2a6eec95b0700b2063a2f8df414037528e90c206648c1c5e49c5e89e741d",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_common.py": "38a29a34663a5d6310571eeff53b5d18a0355e51b9fa58a08033f64fa379a9df",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_common.py": "4e85f5d211c4d3e2188422aa4c0cefa40dcfa14ef32e90894ed3412de1fd1d12",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_data.py": "fe3aa3e023f22f4fe55cd47e7ffa677b58f52a041867ac05620cbd329d913cb6",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_guard.py": "7144fa77ffe61a2a3c92c050234e370ef4dfeb0e31fafb419f752f969737ff1a",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_loss.py": "79e291866bda300ecdc5577c62903d34abeec14f5cc36db05e346f9ffd64c24b",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_queue.py": "a04879cc0169ecb523a88360cddca1e234d91b3adbc20b3ec31d02461ed767eb",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_worker.py": "0f8f35e1422021b2a4f097c2e62267fb935fe4658fd0ee8ccd251609a4062ea7",
|
||||
"/home/frank14f/optane/MeltMind/tools/prepare_phase3_item_c.py": "782c415b1f536eb2b85aa2c9c4a9a0dc278f9b7134e8a83e2a6af8199396c53a",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_custody_common.py": "588c646101be0d50f505f9a12a1c98361aa3437c8c594eecb7321831f717a741",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_custody_statistics.py": "188565801f6901d0192f79aec0889e632501e672a4149d76d1835efbb9b45447",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase3_item_c.py": "3115aa9049974eac86657aa4cee7b5d716f6ee93268555cc86242029ebefce4c",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase2_h2_s2.py": "4235890fd194b7778e937f0be9245b964512df2c75d962ef703e3a988c5b8805",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_guard.py": "6bc7624b0cfe850591f77153cd2034eefdd62f7e63b93211ccdb87eb8aba346c",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_common.py": "308486507ae3589a0e6cc40565f6e60e6d2f63ae74c4d828bdacf33b3b8b8062",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_fixed_io.py": "ea17fdfc7508f3f980e3cdf69534b1eb8ceb8a8e469c32451b2d5e16d0ada575",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_worker.py": "37dfd75fff50b3d65f4ef37f4255536616582fe5f3d01155a0e037325ab5c7cd",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_worker.py": "ecee04b183ef2b93b7b39a447d48c40a5843d3190e48d2d7eb4939811a5db9b0",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_prepare.py": "964fae13ac95e4de4dc50b3c0a80791d86349ce3a9f3c5a6e7dc3cf8ca041fad",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_prepare.py": "3a31af0766f628a80f38e0b7286beb11ca3109188eabc4adfdb2ecb0eef3dc27",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_queue.py": "d82dfb365956e08250758bb4c651a19fbecce92fe209072f5f8ed35970450735",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_launch.py": "480fac25dea5a2006bb84130a4b0a0b253b3040b2150a56c38eada62b2c259e6",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_guard.py": "ae94711adc17a7630f398e8d4120fa3cc8f60979a00122c4cd2dc364c382ced5",
|
||||
"/home/frank14f/optane/MeltMind/docs/phase2-h2-s2-execution-contract-2026-11-17.md": "1e8e5ef31650519737cd6cd247abc0bd04744dd93ffb013eff88866d42f03073"
|
||||
},
|
||||
"verdict": "PASS",
|
||||
"reviewer": "MeltAssist",
|
||||
"scope": "Independent item1b five-fold100epoch training-entry review under Claude1795010666; reused fixed capacity rule, not c entry PASS",
|
||||
"evidence_sha256": "ea3c1fc443e312afeb037a82e2b5786443a454c1d2732716ef2e459869f39bd3",
|
||||
"audit_source_sha256": "0883429df7b4a294c3ddd0635f10b1f9ad073c024a97c8f6c9eff807f7f58609",
|
||||
"findings": [],
|
||||
"passed_checks": [
|
||||
"Exact60code/188source/74local immutable Git files; c52+seed1 51+TrackA27 pins all unchanged.",
|
||||
"Original55ffc6f9 schedule implemented exactly: seed0,5warmup epochs,95cosine decay epochs,1e-4/1e-3/1e-5 endpoints, only finalepoch100; no reuse of original200epoch midpoint.",
|
||||
"Original19inputs/H2/MSE/optimizer/weight decay/clipping/seed0/part-order/learning normalization retained. Training and validation loop AST equals original S2 after declared100epoch and per-fold cost-gate substitutions.",
|
||||
"Original five paired200epoch seed0 model/result/learning-normalization identities verified by metadata and opaque hashes; original historical diagnostic validation timing maxima equal new job bounds. No model/target arrays decoded.",
|
||||
"Per10epoch full-node validation MSE remains diagnostic only. Fold1 same-query-domain component-equal relativeL2 screen is separate and unchanged; delta<=-.001 and abs effect strictly greater than same-domain seed1-seed0, missingreference waits, no newseed.",
|
||||
"Five folds may train before screen per0666. Queue stops at TRAIN_COMPLETE_VALIDATION_SCREEN_PENDING with five frozen models; failed screen archives without test, and test postfit still needs actual separate pre-score review.",
|
||||
"Final test contract preserves552parts/204components,2000paired component bootstrap seed2026111408,95CI, upperdelta strictly<-.001 and abs pooled effect>abs matched seed difference; equality/incomplete fail, development reuse limitations retained.",
|
||||
"Shared adapter AST differs only item1b registration key/cap88560/GPU0..2; admission/registry/transfer/release/validate are same reviewed functions. Mixed c/1b pending reservation blocks overadmission; GPU0 oldpending, GPU3single and CPU/memory rules retained.",
|
||||
"88560 includes post7200GPU/3600CPU, prep600 and per-fold60setup/wait/failure accounting. Actual30000 runaway independent of projected fold cost; total1.5 remaining train+diagnostic cost,120cleanup; phase3<=720000 and TrackA separate.",
|
||||
"Failure/cleanup/late-peak accounting and model freeze inherited unchanged from reviewed c orchestration except declared1b paths/100epoch status. No retries.",
|
||||
"Independent23CPU synthetic tests PASS; measured full childCPU4.113682s; author pre-freeze missingGIB failure and corrected log preserved."
|
||||
],
|
||||
"limitations": [
|
||||
"This PASS binds only45b5df30/d33f036a and full60code map; it is not a scientific result or reuse of c entry PASS.",
|
||||
"No training launch, GPU query, real scientific array/label/model value decode, live source/registry/Git modification. Only own review artifacts and temporary synthetic fixtures written.",
|
||||
"Validation screen/prediction/scoring implementation is not included or approved here. Actual screen, seedreference matching, complete fivefold termination and finalcost gates must hold before separate reviewed test postfit.",
|
||||
"The original corpus has development exposure; one seed difference does not estimate all training randomness.",
|
||||
"Actual shared-card training timings/memory remain runtime evidence; the1.5factor and engineering peaks are limits/projections rather than new1b measurements.",
|
||||
"Publicationffd5c929 differs from implementation45b5df30. Local Git bytes verified; no independent remote raw-byte verification claim."
|
||||
]
|
||||
}
|
||||
File diff suppressed because it is too large.
Load diff
File diff suppressed because it is too large.
Load diff
@@ -0,0 +1,77 @@
|
||||
{
|
||||
"reviewed_commit": "b2a9dfdc4a2c237416452ff6b4c32bdbf34e172e",
|
||||
"contract_sha256": "df80f3b4bbb0d4ee285401ff50b40cfc6bfd06f3302abcd53b41f14a20a204de",
|
||||
"implementation_proof_sha256": "0501275106a0698c9dac6f75554b570bb82f7d4640e76b84f51cf759a8bdadd2",
|
||||
"comparison_contract_sha256": "06556bf1f7bd5a1a79451bd9eed47fc6ec027e7e06bc63c6f490a50dfbee6d20",
|
||||
"code_sha256": {
|
||||
"/home/frank14f/optane/MeltMind/tools/freeze_g1_components.py": "d0cee60d95243ac2c59ef3f2244ba818315718c4e3d1895f2f9f064f1f7f5ecd",
|
||||
"/home/frank14f/optane/MeltMind/tools/gre24_models.py": "53b6fee8b700b9b982b06d29908ce2ce23e0c57512c46dbf74b2aade79e3ac3a",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_review.py": "915617aeb8292d03c654e637bc34a014c254fd178a773f135d1806c4b41372a8",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h1_worker_v2.py": "1543df85b615bb69b79bacd572074a7c77ae19dab110197601573a7e5805ed20",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_data.py": "7e03dd7a94e7bb43593277f183c222473d7cb7aaf2f60e53b45ab8d9499fcb09",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_model.py": "fab2470dd0654ca74c3b7d893f35ad1f3ae6c5c08ffe334ef6312cf6ec87892a",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_pilot_worker.py": "14651ca402e6d02466d8afdb77bc9e62fb2e752f0c0c3d20d98593157d5ebe81",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_common.py": "38a29a34663a5d6310571eeff53b5d18a0355e51b9fa58a08033f64fa379a9df",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_home_freeze_io.py": "bf09204b85ed2a7038acdd202257f618474d1110a9d6d772f105a3ffb334d3ff",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_local_freeze_io.py": "60adf3382577613c5234f5cf30779336b9951b66e90c075cb7cd0adab9cc42dd",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_residual_math.py": "ffbd9624dec79bdaabf95244fe87209055908f31a977e73ea4377045d6f8fe62",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_capacity.py": "3915da37ead45f42fe620cb89bc7394b576f8e113047a4ad78e1bae50ded120c",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_common.py": "d00069d2ca5a346d8260dfb1829bf941f3ea09ebd7d75c1fe377120b553e02d2",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_data.py": "74e35b6268462c999078383e49e551c9da2562390d22977414713811bab164b3",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_guard.py": "e95cd5126f403881aebe137ed3f00bb6173881eea15709952142cb51136ad699",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_model.py": "13000ce6eb7cccfbefdc1710e37a59412be0391ae820b243f0d8fe7a9be4e986",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_queue.py": "0c6f5c9bde84f28bfc829701d97701188f43ed59fa0c5286228ca325cf88235c",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_worker.py": "d49a5e6f900850aabacc5825168b6f886215aff8c494937bea2238539e58f652",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_seed1_v2_common.py": "25ca6be14027369393bd3616209b08f30e25892cccca83300483c5a116e8d75d",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_shared_capacity.py": "7c07b0aaf1bc4e6a980d016defcac326c3bd871134d2583bb4dcd88bb34728f8",
|
||||
"/home/frank14f/optane/MeltMind/tools/prepare_phase3_item_b.py": "ac46c3c6d01d8023444d2f53f956d94d1d295ae95a3435f355c5c6e0f6ccd962",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_gpu_idle_gate.py": "e2b1613e7e6ea398a8d555220a9365a1ad998ab4ef4b2bf8ec7cc4e31f8b4b9f",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_main_memory_guard.py": "e094f5443e11518efe9018215c11718a8e18aa462a977bf43e974b2d0b2aebb1",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/__init__.py": "a0fa1323d3553172da66c5d442c5fc4fcb0d5055c886c776928604550ea97609",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/attention.py": "7714adfccf6a0fc6b52a44aee3a50321a88898bf3d743680a1108c01e1ff8a61",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/blocks.py": "b13e986cea5f1d7f325a1638c1505197b4a105baacff5a9a515f23fcaf28ea75",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/features.py": "3b34ae61cf8012ebd58b7505a73d7aa43d8ee1890ac54f98d7917b67887a45a4",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/network.py": "5137e845d81d1223afc13bf87e03ccba1132a3baa75d5189a8f320b2752be071",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/normalization.py": "7180696d3845799e28d8c31ddef957d075de6a2e46d96a3a0f0df16fc05234a0",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/objectives.py": "5b6527a45f08690055d572335470ce5979e8bb86e1cab00de594f79ae21316b1",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_pilot_train.py": "e9cceead51a5cc092aad23409ad98909c84a62e7e10571e750d1e9be1ad3b948",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_precision_continuation_guard.py": "01235d897c7c03e5ffa109c58b14f5d24d128e8f1ba873e1f76cfb14e02f7a7f",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_recovery_gpu_supervisor.py": "689c0f3179e4fcd362e546d90214829a9e4f8739abdbdf74b560763e5a392f15",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_recovery_monitoring.py": "442c2a9c75e8c00b30a8f2e91cb52d87c466b19a80aea9128b10c021e61b922e",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase3_item_b.py": "bed8e21042971bb688b33c13903a28fab1772b980431a80afd205dbe4350dac2",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_h2_common.py": "79918e7de887d1f0a2de5523156b27d00dc2d85ac930a07efc34d817d904fe65",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_h2_prepare.py": "f657be036dc48fc5936f56c4b8510a4b295d4787af6877cc5ccaad663107be17",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_recovery_monitor.py": "4c423f8b0dff69304d0046c0a9c319bb31aa0c6d1cbaeb3942726b5e6caa0496",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_recovery_supervisor.py": "5ccb2a6eec95b0700b2063a2f8df414037528e90c206648c1c5e49c5e89e741d",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_capacity.py": "771faad0095712fd4fc8131ddcfabaa3901141592d2b055ea2faadf8ea3d3edf",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_common.py": "4e25effe8326ff8c7c3f751590315040209af054b32b0505dc3c6b001589db32",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_guard.py": "08b7d8ab2ac773fab8b749f9f61af259254422128e9634f8b715c81efce2651f",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_queue.py": "31dfd71369c63512c1d6fe176fd22e8db646520c92da533bb5530ca3843c388d",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_schedule.py": "06cb97594c3a87f42f712dcde6cec8c2d90e0b50a89a6be7954bdd2f5c18d5de",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_worker.py": "b781e56c519401bd38af6a522deb24506cba6993d2af008f85ea4e1715f1d9c6",
|
||||
"/home/frank14f/optane/MeltMind/tools/prepare_phase3_item1b.py": "c7bc9f3b7cf4137f52e1efad19dcb7045282d4e9a9c1aae50f066c9e488f7c01",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase3_item1b.py": "b4d78f4384651711fe83ff00eb6719c6c61664699d801ecb34978b9546439da9",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_common.py": "4e85f5d211c4d3e2188422aa4c0cefa40dcfa14ef32e90894ed3412de1fd1d12",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_data.py": "fe3aa3e023f22f4fe55cd47e7ffa677b58f52a041867ac05620cbd329d913cb6",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_guard.py": "7144fa77ffe61a2a3c92c050234e370ef4dfeb0e31fafb419f752f969737ff1a",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_loss.py": "79e291866bda300ecdc5577c62903d34abeec14f5cc36db05e346f9ffd64c24b",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_queue.py": "a04879cc0169ecb523a88360cddca1e234d91b3adbc20b3ec31d02461ed767eb",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_worker.py": "0f8f35e1422021b2a4f097c2e62267fb935fe4658fd0ee8ccd251609a4062ea7",
|
||||
"/home/frank14f/optane/MeltMind/tools/prepare_phase3_item_c.py": "782c415b1f536eb2b85aa2c9c4a9a0dc278f9b7134e8a83e2a6af8199396c53a",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_custody_common.py": "588c646101be0d50f505f9a12a1c98361aa3437c8c594eecb7321831f717a741",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_custody_statistics.py": "188565801f6901d0192f79aec0889e632501e672a4149d76d1835efbb9b45447",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase3_item_c.py": "3115aa9049974eac86657aa4cee7b5d716f6ee93268555cc86242029ebefce4c",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase2_h2_s2.py": "4235890fd194b7778e937f0be9245b964512df2c75d962ef703e3a988c5b8805",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_guard.py": "6bc7624b0cfe850591f77153cd2034eefdd62f7e63b93211ccdb87eb8aba346c",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_common.py": "308486507ae3589a0e6cc40565f6e60e6d2f63ae74c4d828bdacf33b3b8b8062",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_fixed_io.py": "ea17fdfc7508f3f980e3cdf69534b1eb8ceb8a8e469c32451b2d5e16d0ada575",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_worker.py": "37dfd75fff50b3d65f4ef37f4255536616582fe5f3d01155a0e037325ab5c7cd",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_worker.py": "ecee04b183ef2b93b7b39a447d48c40a5843d3190e48d2d7eb4939811a5db9b0",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_prepare.py": "964fae13ac95e4de4dc50b3c0a80791d86349ce3a9f3c5a6e7dc3cf8ca041fad",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_prepare.py": "3a31af0766f628a80f38e0b7286beb11ca3109188eabc4adfdb2ecb0eef3dc27",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_queue.py": "d82dfb365956e08250758bb4c651a19fbecce92fe209072f5f8ed35970450735",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_launch.py": "480fac25dea5a2006bb84130a4b0a0b253b3040b2150a56c38eada62b2c259e6",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_guard.py": "ae94711adc17a7630f398e8d4120fa3cc8f60979a00122c4cd2dc364c382ced5",
|
||||
"/home/frank14f/optane/MeltMind/docs/phase2-h2-s2-execution-contract-2026-11-17.md": "1e8e5ef31650519737cd6cd247abc0bd04744dd93ffb013eff88866d42f03073"
|
||||
}
|
||||
}
|
||||
@@ -0,0 +1,4 @@
|
||||
{
|
||||
"contract_sha256": "df80f3b4bbb0d4ee285401ff50b40cfc6bfd06f3302abcd53b41f14a20a204de",
|
||||
"proof_sha256": "0501275106a0698c9dac6f75554b570bb82f7d4640e76b84f51cf759a8bdadd2"
|
||||
}
|
||||
@@ -0,0 +1,169 @@
|
||||
{
|
||||
"commit": "b2a9dfdc4a2c237416452ff6b4c32bdbf34e172e",
|
||||
"files": {
|
||||
"/home/frank14f/optane/MeltMind/tools/freeze_g1_components.py": "d0cee60d95243ac2c59ef3f2244ba818315718c4e3d1895f2f9f064f1f7f5ecd",
|
||||
"/home/frank14f/optane/MeltMind/tools/gre24_models.py": "53b6fee8b700b9b982b06d29908ce2ce23e0c57512c46dbf74b2aade79e3ac3a",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_review.py": "915617aeb8292d03c654e637bc34a014c254fd178a773f135d1806c4b41372a8",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h1_worker_v2.py": "1543df85b615bb69b79bacd572074a7c77ae19dab110197601573a7e5805ed20",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_data.py": "7e03dd7a94e7bb43593277f183c222473d7cb7aaf2f60e53b45ab8d9499fcb09",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_model.py": "fab2470dd0654ca74c3b7d893f35ad1f3ae6c5c08ffe334ef6312cf6ec87892a",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_pilot_worker.py": "14651ca402e6d02466d8afdb77bc9e62fb2e752f0c0c3d20d98593157d5ebe81",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_common.py": "38a29a34663a5d6310571eeff53b5d18a0355e51b9fa58a08033f64fa379a9df",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_home_freeze_io.py": "bf09204b85ed2a7038acdd202257f618474d1110a9d6d772f105a3ffb334d3ff",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_local_freeze_io.py": "60adf3382577613c5234f5cf30779336b9951b66e90c075cb7cd0adab9cc42dd",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_residual_math.py": "ffbd9624dec79bdaabf95244fe87209055908f31a977e73ea4377045d6f8fe62",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_capacity.py": "3915da37ead45f42fe620cb89bc7394b576f8e113047a4ad78e1bae50ded120c",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_common.py": "d00069d2ca5a346d8260dfb1829bf941f3ea09ebd7d75c1fe377120b553e02d2",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_data.py": "74e35b6268462c999078383e49e551c9da2562390d22977414713811bab164b3",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_guard.py": "e95cd5126f403881aebe137ed3f00bb6173881eea15709952142cb51136ad699",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_model.py": "13000ce6eb7cccfbefdc1710e37a59412be0391ae820b243f0d8fe7a9be4e986",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_queue.py": "0c6f5c9bde84f28bfc829701d97701188f43ed59fa0c5286228ca325cf88235c",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_worker.py": "d49a5e6f900850aabacc5825168b6f886215aff8c494937bea2238539e58f652",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_seed1_v2_common.py": "25ca6be14027369393bd3616209b08f30e25892cccca83300483c5a116e8d75d",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_shared_capacity.py": "7c07b0aaf1bc4e6a980d016defcac326c3bd871134d2583bb4dcd88bb34728f8",
|
||||
"/home/frank14f/optane/MeltMind/tools/prepare_phase3_item_b.py": "ac46c3c6d01d8023444d2f53f956d94d1d295ae95a3435f355c5c6e0f6ccd962",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_gpu_idle_gate.py": "e2b1613e7e6ea398a8d555220a9365a1ad998ab4ef4b2bf8ec7cc4e31f8b4b9f",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_main_memory_guard.py": "e094f5443e11518efe9018215c11718a8e18aa462a977bf43e974b2d0b2aebb1",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/__init__.py": "a0fa1323d3553172da66c5d442c5fc4fcb0d5055c886c776928604550ea97609",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/attention.py": "7714adfccf6a0fc6b52a44aee3a50321a88898bf3d743680a1108c01e1ff8a61",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/blocks.py": "b13e986cea5f1d7f325a1638c1505197b4a105baacff5a9a515f23fcaf28ea75",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/features.py": "3b34ae61cf8012ebd58b7505a73d7aa43d8ee1890ac54f98d7917b67887a45a4",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/network.py": "5137e845d81d1223afc13bf87e03ccba1132a3baa75d5189a8f320b2752be071",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/normalization.py": "7180696d3845799e28d8c31ddef957d075de6a2e46d96a3a0f0df16fc05234a0",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/objectives.py": "5b6527a45f08690055d572335470ce5979e8bb86e1cab00de594f79ae21316b1",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_pilot_train.py": "e9cceead51a5cc092aad23409ad98909c84a62e7e10571e750d1e9be1ad3b948",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_precision_continuation_guard.py": "01235d897c7c03e5ffa109c58b14f5d24d128e8f1ba873e1f76cfb14e02f7a7f",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_recovery_gpu_supervisor.py": "689c0f3179e4fcd362e546d90214829a9e4f8739abdbdf74b560763e5a392f15",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_recovery_monitoring.py": "442c2a9c75e8c00b30a8f2e91cb52d87c466b19a80aea9128b10c021e61b922e",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase3_item_b.py": "bed8e21042971bb688b33c13903a28fab1772b980431a80afd205dbe4350dac2",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_h2_common.py": "79918e7de887d1f0a2de5523156b27d00dc2d85ac930a07efc34d817d904fe65",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_h2_prepare.py": "f657be036dc48fc5936f56c4b8510a4b295d4787af6877cc5ccaad663107be17",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_recovery_monitor.py": "4c423f8b0dff69304d0046c0a9c319bb31aa0c6d1cbaeb3942726b5e6caa0496",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_recovery_supervisor.py": "5ccb2a6eec95b0700b2063a2f8df414037528e90c206648c1c5e49c5e89e741d",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_capacity.py": "771faad0095712fd4fc8131ddcfabaa3901141592d2b055ea2faadf8ea3d3edf",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_common.py": "4e25effe8326ff8c7c3f751590315040209af054b32b0505dc3c6b001589db32",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_guard.py": "08b7d8ab2ac773fab8b749f9f61af259254422128e9634f8b715c81efce2651f",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_queue.py": "31dfd71369c63512c1d6fe176fd22e8db646520c92da533bb5530ca3843c388d",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_schedule.py": "06cb97594c3a87f42f712dcde6cec8c2d90e0b50a89a6be7954bdd2f5c18d5de",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_worker.py": "b781e56c519401bd38af6a522deb24506cba6993d2af008f85ea4e1715f1d9c6",
|
||||
"/home/frank14f/optane/MeltMind/tools/prepare_phase3_item1b.py": "c7bc9f3b7cf4137f52e1efad19dcb7045282d4e9a9c1aae50f066c9e488f7c01",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase3_item1b.py": "b4d78f4384651711fe83ff00eb6719c6c61664699d801ecb34978b9546439da9",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_common.py": "4e85f5d211c4d3e2188422aa4c0cefa40dcfa14ef32e90894ed3412de1fd1d12",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_data.py": "fe3aa3e023f22f4fe55cd47e7ffa677b58f52a041867ac05620cbd329d913cb6",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_guard.py": "7144fa77ffe61a2a3c92c050234e370ef4dfeb0e31fafb419f752f969737ff1a",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_loss.py": "79e291866bda300ecdc5577c62903d34abeec14f5cc36db05e346f9ffd64c24b",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_queue.py": "a04879cc0169ecb523a88360cddca1e234d91b3adbc20b3ec31d02461ed767eb",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_worker.py": "0f8f35e1422021b2a4f097c2e62267fb935fe4658fd0ee8ccd251609a4062ea7",
|
||||
"/home/frank14f/optane/MeltMind/tools/prepare_phase3_item_c.py": "782c415b1f536eb2b85aa2c9c4a9a0dc278f9b7134e8a83e2a6af8199396c53a",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_custody_common.py": "588c646101be0d50f505f9a12a1c98361aa3437c8c594eecb7321831f717a741",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_custody_statistics.py": "188565801f6901d0192f79aec0889e632501e672a4149d76d1835efbb9b45447",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase3_item_c.py": "3115aa9049974eac86657aa4cee7b5d716f6ee93268555cc86242029ebefce4c",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase2_h2_s2.py": "4235890fd194b7778e937f0be9245b964512df2c75d962ef703e3a988c5b8805",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_guard.py": "6bc7624b0cfe850591f77153cd2034eefdd62f7e63b93211ccdb87eb8aba346c",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_common.py": "308486507ae3589a0e6cc40565f6e60e6d2f63ae74c4d828bdacf33b3b8b8062",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_fixed_io.py": "ea17fdfc7508f3f980e3cdf69534b1eb8ceb8a8e469c32451b2d5e16d0ada575",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_worker.py": "37dfd75fff50b3d65f4ef37f4255536616582fe5f3d01155a0e037325ab5c7cd",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_worker.py": "ecee04b183ef2b93b7b39a447d48c40a5843d3190e48d2d7eb4939811a5db9b0",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_prepare.py": "964fae13ac95e4de4dc50b3c0a80791d86349ce3a9f3c5a6e7dc3cf8ca041fad",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_prepare.py": "3a31af0766f628a80f38e0b7286beb11ca3109188eabc4adfdb2ecb0eef3dc27",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_queue.py": "d82dfb365956e08250758bb4c651a19fbecce92fe209072f5f8ed35970450735",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_launch.py": "480fac25dea5a2006bb84130a4b0a0b253b3040b2150a56c38eada62b2c259e6",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_guard.py": "ae94711adc17a7630f398e8d4120fa3cc8f60979a00122c4cd2dc364c382ced5",
|
||||
"/home/frank14f/optane/MeltMind/docs/phase2-h2-s2-execution-contract-2026-11-17.md": "1e8e5ef31650519737cd6cd247abc0bd04744dd93ffb013eff88866d42f03073",
|
||||
"/home/frank14f/MeltMind-runs/phase3-item-b-2026-11-18/contract.json": "df80f3b4bbb0d4ee285401ff50b40cfc6bfd06f3302abcd53b41f14a20a204de",
|
||||
"/home/frank14f/MeltMind-runs/phase3-item-b-2026-11-18/comparison-contract.json": "06556bf1f7bd5a1a79451bd9eed47fc6ec027e7e06bc63c6f490a50dfbee6d20",
|
||||
"/home/frank14f/MeltMind-runs/phase3-item-b-2026-11-18/preparation-audit.json": "647340e7cf1b0cf47bc4638d91816430419bd63ea7b246067d7a909b463e21e4",
|
||||
"/home/frank14f/optane/MeltMind/docs/phase3-item-b-entry-2026-11-18.md": "9fd1801566b245079430d078b16574f05de9c9f287a85afaf7f8d3197c3194c1",
|
||||
"/home/frank14f/optane/MeltMind/references/2026-11-18/phase3-item-b/mold-regional-context-spec.md": "de120e9dc5527c01af5247a27c8cf529985c206821553bacb01be6071f2cdab5",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/phase3-item-c-launch-2026-11-18/direct-user-authorization.json": "84c315fda1d3c35c36603a7f92443af2c5b91cc03a6fe848edf59f18dc810a81",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/phase3-item-b-acceptance-2026-11-18/attempt1.log": "b562ac199565149af9d5f54f01a5c91427c17ad89f1ae20d28936a3efd622c6c",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/phase3-item-b-acceptance-2026-11-18/acceptance-attempt2.json": "3daee9e7153fdf08cceae27c38c804c95bb0582b275d15632ac952068c3d9182",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/phase3-item-b-acceptance-2026-11-18/acceptance-attempt1.json": "08e6c837254a190ba94a948723a683a43d3197ba84d0b92589db9ae8110eeb35",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/phase3-item-b-acceptance-2026-11-18/attempt2.log": "d431c402bc18a231d9629306ade0e1fa6af2ed378a0ba3230cde45096f1ff54d",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/phase3-item-b-acceptance-2026-11-18/pre-freeze-pin-audit.json": "cda08b576f8c88767211e3e08e2c7f7b296d0868bd5a25fda94746234d849e59"
|
||||
},
|
||||
"git_paths": {
|
||||
"/home/frank14f/optane/MeltMind/tools/freeze_g1_components.py": "tools/freeze_g1_components.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/gre24_models.py": "tools/gre24_models.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_gate_outcome_review.py": "tools/phase2_gate_outcome_review.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h1_worker_v2.py": "tools/phase2_h1_worker_v2.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_data.py": "tools/phase2_h2_data.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_model.py": "tools/phase2_h2_model.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_pilot_worker.py": "tools/phase2_h2_pilot_worker.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_common.py": "tools/phase2_h2_s1_common.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_home_freeze_io.py": "tools/phase2_home_freeze_io.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_local_freeze_io.py": "tools/phase2_local_freeze_io.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_residual_math.py": "tools/phase2_residual_math.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_capacity.py": "tools/phase3_item_b_capacity.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_common.py": "tools/phase3_item_b_common.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_data.py": "tools/phase3_item_b_data.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_guard.py": "tools/phase3_item_b_guard.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_model.py": "tools/phase3_item_b_model.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_queue.py": "tools/phase3_item_b_queue.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_b_worker.py": "tools/phase3_item_b_worker.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_seed1_v2_common.py": "tools/phase3_seed1_v2_common.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_shared_capacity.py": "tools/phase3_shared_capacity.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/prepare_phase3_item_b.py": "tools/prepare_phase3_item_b.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_gpu_idle_gate.py": "tools/s2_gpu_idle_gate.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_main_memory_guard.py": "tools/s2_main_memory_guard.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/__init__.py": "tools/s2_models/__init__.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/attention.py": "tools/s2_models/attention.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/blocks.py": "tools/s2_models/blocks.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/features.py": "tools/s2_models/features.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/network.py": "tools/s2_models/network.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/normalization.py": "tools/s2_models/normalization.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_models/objectives.py": "tools/s2_models/objectives.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_pilot_train.py": "tools/s2_pilot_train.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_precision_continuation_guard.py": "tools/s2_precision_continuation_guard.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_recovery_gpu_supervisor.py": "tools/s2_recovery_gpu_supervisor.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_recovery_monitoring.py": "tools/s2_recovery_monitoring.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase3_item_b.py": "tools/test_phase3_item_b.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_h2_common.py": "tools/track_a_h2_common.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_h2_prepare.py": "tools/track_a_h2_prepare.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_recovery_monitor.py": "tools/track_a_recovery_monitor.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/track_a_recovery_supervisor.py": "tools/track_a_recovery_supervisor.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_capacity.py": "tools/phase3_item1b_capacity.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_common.py": "tools/phase3_item1b_common.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_guard.py": "tools/phase3_item1b_guard.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_queue.py": "tools/phase3_item1b_queue.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_schedule.py": "tools/phase3_item1b_schedule.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item1b_worker.py": "tools/phase3_item1b_worker.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/prepare_phase3_item1b.py": "tools/prepare_phase3_item1b.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase3_item1b.py": "tools/test_phase3_item1b.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_common.py": "tools/phase3_item_c_common.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_data.py": "tools/phase3_item_c_data.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_guard.py": "tools/phase3_item_c_guard.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_loss.py": "tools/phase3_item_c_loss.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_queue.py": "tools/phase3_item_c_queue.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase3_item_c_worker.py": "tools/phase3_item_c_worker.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/prepare_phase3_item_c.py": "tools/prepare_phase3_item_c.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_custody_common.py": "tools/s2_custody_common.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/s2_custody_statistics.py": "tools/s2_custody_statistics.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase3_item_c.py": "tools/test_phase3_item_c.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/test_phase2_h2_s2.py": "tools/test_phase2_h2_s2.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_guard.py": "tools/phase2_h2_s2_guard.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_common.py": "tools/phase2_h2_s2_common.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_fixed_io.py": "tools/phase2_fixed_io.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_worker.py": "tools/phase2_h2_s2_worker.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_worker.py": "tools/phase2_h2_s1_worker.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_prepare.py": "tools/phase2_h2_s2_prepare.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_prepare.py": "tools/phase2_h2_prepare.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s2_queue.py": "tools/phase2_h2_s2_queue.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_launch.py": "tools/phase2_h2_s1_launch.py",
|
||||
"/home/frank14f/optane/MeltMind/tools/phase2_h2_s1_guard.py": "tools/phase2_h2_s1_guard.py",
|
||||
"/home/frank14f/optane/MeltMind/docs/phase2-h2-s2-execution-contract-2026-11-17.md": "docs/phase2-h2-s2-execution-contract-2026-11-17.md",
|
||||
"/home/frank14f/MeltMind-runs/phase3-item-b-2026-11-18/contract.json": "home-runs/phase3-item-b-2026-11-18/contract.json",
|
||||
"/home/frank14f/MeltMind-runs/phase3-item-b-2026-11-18/comparison-contract.json": "home-runs/phase3-item-b-2026-11-18/comparison-contract.json",
|
||||
"/home/frank14f/MeltMind-runs/phase3-item-b-2026-11-18/preparation-audit.json": "home-runs/phase3-item-b-2026-11-18/preparation-audit.json",
|
||||
"/home/frank14f/optane/MeltMind/docs/phase3-item-b-entry-2026-11-18.md": "docs/phase3-item-b-entry-2026-11-18.md",
|
||||
"/home/frank14f/optane/MeltMind/references/2026-11-18/phase3-item-b/mold-regional-context-spec.md": "references/2026-11-18/phase3-item-b/mold-regional-context-spec.md",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/phase3-item-c-launch-2026-11-18/direct-user-authorization.json": "artifacts/phase3-item-c-launch-2026-11-18/direct-user-authorization.json",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/phase3-item-b-acceptance-2026-11-18/attempt1.log": "artifacts/phase3-item-b-acceptance-2026-11-18/attempt1.log",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/phase3-item-b-acceptance-2026-11-18/acceptance-attempt2.json": "artifacts/phase3-item-b-acceptance-2026-11-18/acceptance-attempt2.json",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/phase3-item-b-acceptance-2026-11-18/acceptance-attempt1.json": "artifacts/phase3-item-b-acceptance-2026-11-18/acceptance-attempt1.json",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/phase3-item-b-acceptance-2026-11-18/attempt2.log": "artifacts/phase3-item-b-acceptance-2026-11-18/attempt2.log",
|
||||
"/home/frank14f/optane/MeltMind/artifacts/phase3-item-b-acceptance-2026-11-18/pre-freeze-pin-audit.json": "artifacts/phase3-item-b-acceptance-2026-11-18/pre-freeze-pin-audit.json"
|
||||
},
|
||||
"remote_published": false,
|
||||
"local_repository": "/tmp/meltmind-phase2-local-fixed-20261117"
|
||||
}
|
||||
@@ -0,0 +1,15 @@
|
||||
{
|
||||
"status": "PREPARED_PENDING_ASSIST_REVIEW",
|
||||
"contract_sha256": "df80f3b4bbb0d4ee285401ff50b40cfc6bfd06f3302abcd53b41f14a20a204de",
|
||||
"code_files": 69,
|
||||
"source_files": 1300,
|
||||
"context_parts": 552,
|
||||
"valid_parts": 510,
|
||||
"scientific_missing_parts": 42,
|
||||
"opaque_cache_hashes_only": true,
|
||||
"metadata_cpu_seconds": 2.210258901,
|
||||
"labels_decoded": false,
|
||||
"arrays_decoded": false,
|
||||
"models_loaded": false,
|
||||
"new_fits_started": false
|
||||
}
|
||||
@@ -0,0 +1,23 @@
|
||||
# H2 后续 B|原网格网络还能否利用八区背景?
|
||||
|
||||
MoldAstra · 2026-11-17 · 事前规格。实施次序 A 种子、B 区域、C 末填。仅在 H2-S2 确认后由 Claude 排期。主对照是原版单模型 H2 seed 0。A 的集成结果不参与本项输入或基线。
|
||||
|
||||
## 证据与待检假设
|
||||
|
||||
B13-E1b 在旧五折树模型上把场误差从约 7.08% 降到 6.97%。H2 有四面体局部消息和每个物理连通分量的全局池化。粗区均值可能为其提供有限层数难以学到的中程信息,也可能已被现有网络充分利用。这必须在相同五折 H2 上做配对。树模型五折与 H2 首折的收益不可直接排序。[B13 几何定义](https://drive.google.com/file/d/190x6wRmrojGwIlVC4kldcXbyH_abppgt/view) · [E1b 全件含缺失的裁量](https://arcticnimbus.slack.com/archives/C0C78UFKGRM/p1794905109925079) · [H2 架构](https://drive.google.com/file/d/13_ckwlmo9Qfb5sFzs5pwjla8_4scccMR/view)
|
||||
|
||||
## 唯一干预与缺失处理
|
||||
|
||||
读取已冻结 G 缓存。八个区由共面四面体邻接图产生;每个单元先取四节点的 F 和 Dw* 均值,再取区域内体积加权均值。界面节点按其入射单元体积混合所属区的两个均值。保留原网格原 ID,区号本身不进网络。F 是求解前的严格体积分位,Dw* 为冻结的归一化壁距列。常数再现、区号重排不变、集总质量加权全件积分守恒均由读出定义推出;后者不等于熔体通量守恒。
|
||||
|
||||
原 19 列追加两个数值列和一个两列共同有效的 0/1 指示,共 22 列。只在该折学习池原监督节点的有效行上按原 H2 节点权口径定两列均值与尺度;零方差取尺度 1。按原 B13 共同缺失约定,无效件两列标准化后置零,指示为零;有效行即使刚好等于均值,指示仍为一。若原缓存出现非预期的单列缺失或 ID/网格身份错误,作为输入不一致披露,不将错误转换为“几何无效”。全名单保留。仅新列进入原本接收节点特征的编码和解码层;旧列权重与原 seed 0 初值对齐,新列初始权重为零,保留所有共享参数初值及洗牌顺序。原 FP64 的 cF 跳接不变。宽 64 的两个首层各增 3×64 个权重,共 384 个。指标比较的是两列背景和缺失信息整体的净增量。
|
||||
|
||||
## 评分与判决
|
||||
|
||||
只拟合一个 H2+G 配方,seed 0 五折、100% 各自学习池,原训练轮数、时间目标、损失、优化器和最终检查点。对照用原 S2 同 fold、同 raw ID 的 H2 场。552 件/204 组件上的主量为每件 relative L2 新减旧,组件内先平均,再跨组件等权;成对组件 bootstrap 2,000 次、seed 2026111408、双侧 95%。完整名单且上界 <−0.001,记“八区输入达到预设场收益”,否则不推荐将它设为默认输入。该 0.10 个百分点为研究成本门槛。旧库上的一次受限比较不能证明任意新家族的性能。
|
||||
|
||||
同时报告 G 有效与无效的件/组件数、两个子集的配对差值、各折结果、末点/Rg、τ、负时间和端到端时间。若收益集中在无效件,不能归因于八区的空间信息。即使末点恶化也照报;没有独立的末点通过判定。不因 G 无效挑除难件、再挑切分种子或调 K。
|
||||
|
||||
## 成本
|
||||
|
||||
五次新增拟合。按 fold 0 累计 5.24 GPU-h 加 25% 估算约 32.75 GPU-h,另列新列搬运、前向及评分;CPU 主要为读取同身份 G、按 ID 回填,随节点/入射单元线性增长。每百万节点新增三列 FP32 输入约 12 MB,不含激活及缓存。部署比原 H2 多一次 G 准备,缓存失效或新 CAD 上要载明分区 CPU 用时;零缓存的端到端时延单列。区分已有缓存与从头分区。预算估算以 S2 各折实账更新,不占现有训练额度。[成本依据](https://arcticnimbus.slack.com/archives/C0C78UFKGRM/p1794923791467409)
|
||||
@@ -0,0 +1,65 @@
|
||||
"""b registration wrapper; unchanged reviewed shared admission and leases.
|
||||
|
||||
Only entry key/cap and eligibleGPU0..2 differ from c's registration/acquire.
|
||||
The live c module is imported without mutation, never patched or rewritten.
|
||||
"""
|
||||
import os,time,uuid
|
||||
from pathlib import Path
|
||||
import psutil
|
||||
from phase3_shared_capacity import (registry,PHASE_RESERVES,process_identity,alive,descendants,
|
||||
host_snapshot,admission,GPUS,GIB,transfer,release,validate_lease)
|
||||
|
||||
def register_entry(root, digest):
|
||||
with registry() as d:
|
||||
if 'b' in d['entries']:
|
||||
raise FileExistsError('Sole item b entry already registered; no resume')
|
||||
if sum(PHASE_RESERVES.values()) > 720000:
|
||||
raise RuntimeError('Whole phase3 reserved cap')
|
||||
d['entries']['b'] = dict(root=str(root), contract_sha256=digest, owner=process_identity(os.getpid()),
|
||||
cap=178200, phase_reservations=PHASE_RESERVES, started_unix=time.time())
|
||||
|
||||
|
||||
def acquire(root, fold, digest):
|
||||
with registry() as d:
|
||||
entry = d['entries']['b']
|
||||
if entry['root'] != str(root) or entry['contract_sha256'] != digest or not alive(entry['owner']):
|
||||
raise PermissionError('Entry owner/identity')
|
||||
leases = []
|
||||
for l in d['leases'].values():
|
||||
# Normal guard exit can race this observation. Block admission until
|
||||
# its owning coordinator collects it; never reclaim or kill peers.
|
||||
if not alive(l['owner']):
|
||||
return None, dict(reason='STALE_LEASE_AWAIT_OWNER_COLLECTION',token=l['token'])
|
||||
if l['state']=='RESERVED':
|
||||
# owner is a queue with other children, not this job's guard.
|
||||
# One independent slot and full VRAM/RSS remain reserved.
|
||||
leases.append(dict(l,observed_pids=[],observed_rss=0))
|
||||
continue
|
||||
if l['state']!='GUARD_BOUND':
|
||||
raise ValueError('Unknown lease state')
|
||||
try:
|
||||
ids = descendants(l['owner'])
|
||||
except (psutil.NoSuchProcess,RuntimeError):
|
||||
return None, dict(reason='LEASE_EXIT_RACE_AWAIT_OWNER_COLLECTION',token=l['token'])
|
||||
rss=0
|
||||
for pid in ids:
|
||||
try:rss+=psutil.Process(pid).memory_info().rss
|
||||
except psutil.NoSuchProcess:pass
|
||||
leases.append(dict(l, observed_pids=list(ids), observed_rss=rss))
|
||||
s = host_snapshot() # fresh inside inter-entry lock, before making reservation
|
||||
used_cpu = {x for l in leases for x in l['worker_cpus']}
|
||||
cpus = [x for x in range(16,26) if x not in used_cpu and x in os.sched_getaffinity(0)]
|
||||
if not cpus:
|
||||
return None, dict(reason='CPU_RESERVATIONS', snapshot=s)
|
||||
for gpu in GPUS[:3]:
|
||||
ok, reason = admission(s, leases, gpu)
|
||||
if not ok:
|
||||
continue
|
||||
token = uuid.uuid4().hex
|
||||
queue_owner=process_identity(os.getpid())
|
||||
l = dict(token=token, state='RESERVED',root=str(root), fold=fold, contract_sha256=digest, gpu_uuid=gpu,
|
||||
worker_cpus=[cpus[0]], owner=queue_owner,queue_owner=queue_owner,
|
||||
vram_reserve=4*GIB, rss_reserve=4*GIB, created_unix=time.time())
|
||||
d['leases'][token] = l
|
||||
return l, dict(reason='ADMITTED', snapshot=s)
|
||||
return None, dict(reason='NO_CAPACITY', snapshot=s)
|
||||
@@ -0,0 +1,112 @@
|
||||
"""Item b immutable scope, assignments and conservative total-cost projection."""
|
||||
import json
|
||||
import math
|
||||
import statistics
|
||||
from pathlib import Path
|
||||
from phase2_h1_worker_v2 import checked, sha
|
||||
from phase2_home_freeze_io import verify
|
||||
from phase3_seed1_v2_common import fold_view
|
||||
|
||||
ROOT = Path('/home/frank14f/MeltMind-runs/phase3-item-b-2026-11-18')
|
||||
|
||||
|
||||
def read(p):
|
||||
return json.loads(Path(p).read_text())
|
||||
|
||||
|
||||
def config(root):
|
||||
if root.resolve() != ROOT:
|
||||
raise PermissionError('Sole item b root')
|
||||
c = read(root/'contract.json')
|
||||
verify(root, 'implementation')
|
||||
expected = dict(approval='1795011607.907609', seed=0, epochs=200, new_fits=5,
|
||||
gpu_cap=178200, postfit_gpu_reserve=7200, postfit_cpu_cap=3600,
|
||||
phase3_gpu_cap=720000, no_retry=True)
|
||||
if any(c.get(k) != v for k,v in expected.items()):
|
||||
raise PermissionError('Fixed item_b scope/budgets')
|
||||
for p,h in {**c['source_sha256'], **c['code_sha256']}.items():
|
||||
checked(p,h)
|
||||
pilot = read(c['pilot_contract'])
|
||||
import numpy as np
|
||||
import torch
|
||||
if np.__version__ != pilot['versions']['numpy'] or torch.__version__ != pilot['versions']['torch']:
|
||||
raise ValueError('Original environment versions')
|
||||
if [j['id'] for j in c['jobs']] != list(range(5)):
|
||||
raise ValueError('Exactly five folds')
|
||||
tests = []
|
||||
for job in c['jobs']:
|
||||
f = fold_view(pilot, job['id'])['folds'][0]
|
||||
tests.extend(f['test'])
|
||||
if job['target_steps'] != 200*len(f['learning']):
|
||||
raise ValueError('Full fixed epochs')
|
||||
for k in ('learning','validation','test','c','clock_sha256','learning_nodes'):
|
||||
if job[k] != f[k]:
|
||||
raise ValueError('Original fold identity changed')
|
||||
if len(tests) != 552 or len(set(tests)) != 552:
|
||||
raise ValueError('Original552 roster')
|
||||
rows=read(c['query_roster_contract'])['rows']
|
||||
if len(rows)!=552 or {r['case'] for r in rows}!=set(tests) or len({r['component_id'] for r in rows})!=204:
|
||||
raise ValueError('552 parts /204 components query roster')
|
||||
for r in rows:
|
||||
p=pilot['parts'][r['case']]
|
||||
if r['v5_cache']['output_sha256']!=p['output_sha256'] or r['nodes']!=p['nodes'] or r['component_id']!=p['clock_group']:
|
||||
raise ValueError('Learning query/geometry/cache identity')
|
||||
if c['inputs']!=22 or c['added_parameters']!=384 or c['budget_approval']!='1795007580.895539' or not c['decision_requires_seed_reference']:
|
||||
raise PermissionError('Fixed regional context scope')
|
||||
checked(c['comparison_contract'], c['comparison_contract_sha256'])
|
||||
return c,pilot
|
||||
|
||||
|
||||
def assignment(root,c,index,digest):
|
||||
a = read(checked(root/'jobs'/str(index)/'assignment.json',digest))
|
||||
if a['id'] != index or a['contract_sha256'] != sha(root/'contract.json'):
|
||||
raise PermissionError('Assignment identity')
|
||||
return dict(c['jobs'][index], **{k:a[k] for k in ('gpu_uuid','worker_cpus','token','queue_identity')})
|
||||
|
||||
|
||||
def first_epoch_release(root,index):
|
||||
if (root/'global-stop.json').exists():
|
||||
raise RuntimeError('Global stop')
|
||||
p = root/'jobs'/str(index)/'first-epoch-projection.json'
|
||||
if not p.exists():
|
||||
return False
|
||||
d = read(p)
|
||||
if d['status'] != 'PASS' or d['contract_sha256'] != sha(root/'contract.json'):
|
||||
raise RuntimeError('First-epoch gate not passed')
|
||||
return True
|
||||
|
||||
|
||||
def project(c, charges, histories, terminal):
|
||||
if set(charges) != set(range(5)) or set(histories)-set(charges) or set(terminal)-set(charges):
|
||||
raise ValueError('Cost ledger fold identity')
|
||||
values = {}
|
||||
for i,v in charges.items():
|
||||
if not math.isfinite(v) or v < 0:
|
||||
raise ValueError('Invalid actual cost')
|
||||
if i in terminal:
|
||||
values[i] = v
|
||||
elif histories.get(i):
|
||||
rows = histories[i]
|
||||
e = max(r['epoch'] for r in rows)
|
||||
if sorted(r['epoch'] for r in rows) != list(range(1,e+1)) or e > 200:
|
||||
raise ValueError('Incomplete epoch history')
|
||||
t = max(r['train_epoch_seconds'] for r in rows)
|
||||
if not math.isfinite(t) or t <= 0:
|
||||
raise ValueError('Invalid measured epoch')
|
||||
# Factor1.5 applies to remaining training, never to already charged cost.
|
||||
validation=max(c['jobs'][i]['validation_seconds_bound'],max(r.get('validation_seconds',0) for r in rows))
|
||||
if not math.isfinite(validation) or validation<0:raise ValueError('Invalid validation timing')
|
||||
values[i] = v + 1.5*((200-e)*t+(20-e//10)*validation) + 120
|
||||
initial = (c['gpu_cap']-c['postfit_gpu_reserve']-c['preparation_charge_gpu_seconds'])/5
|
||||
median = statistics.median(values.values()) if values else initial
|
||||
for i in charges:
|
||||
if i not in values:
|
||||
values[i] = max(charges[i]+120, initial, median)
|
||||
actual = c['preparation_charge_gpu_seconds']+sum(charges.values())
|
||||
projected = c['preparation_charge_gpu_seconds']+sum(values.values())+c['postfit_gpu_reserve']
|
||||
ok = projected <= c['gpu_cap'] and actual+c['postfit_gpu_reserve'] <= c['gpu_cap']
|
||||
ok &= all(charges[i] < c['runaway_gpu_seconds'] for i in charges if i not in terminal)
|
||||
return dict(status='PASS' if ok else 'STOP_BUDGET', projected_gpu_seconds=projected,
|
||||
actual_training_and_preparation_gpu_seconds=actual,
|
||||
folds={str(i):dict(charged=charges[i], projected=values[i]) for i in charges},
|
||||
postfit_gpu_reserve=c['postfit_gpu_reserve'], safety_factor=1.5)
|
||||
@@ -0,0 +1,71 @@
|
||||
"""Reuse frozen B13 readout; learning-only normalization, joint missingness, no partition."""
|
||||
import json,time
|
||||
from pathlib import Path
|
||||
import numpy as np
|
||||
import torch
|
||||
from phase2_h1_worker_v2 import checked,save
|
||||
from phase2_h2_data import standardization as original_standardization,load_part as original_load
|
||||
|
||||
|
||||
def valid_rows(extra,status,nodes):
|
||||
x=np.asarray(extra)
|
||||
if x.shape!=(nodes,2) or np.isinf(x).any():
|
||||
raise ValueError('Context shape/infinity input inconsistency')
|
||||
finite=np.isfinite(x)
|
||||
if np.any(finite[:,0]!=finite[:,1]):
|
||||
raise ValueError('Unexpected single-column missingness')
|
||||
if status=='GEOMETRY_COMPLETE' and finite.all():return np.ones(nodes,dtype=bool)
|
||||
if status=='SCIENTIFIC_MISSING' and np.isnan(x).all():return np.zeros(nodes,dtype=bool)
|
||||
raise ValueError('Frozen whole-part context status inconsistent with array')
|
||||
|
||||
|
||||
def read_context(old,c,case):
|
||||
p=c['context'][case];r=old['parts'][case]
|
||||
if p['nodes']!=r['nodes'] or p['raw_ids_sha256']!=r['output_sha256']['raw_ids.npy'] or p['features_sha256']!=r['output_sha256']['features.npy']:
|
||||
raise ValueError('Context original node/feature identity')
|
||||
result=json.loads(Path(checked(p['result'],p['result_sha256'])).read_text())
|
||||
if result['case']!=case or result['status']!=p['status'] or result['extra_sha256']!=p['extra_sha256']:
|
||||
raise ValueError('Context result identity')
|
||||
x=np.load(checked(p['extra'],p['extra_sha256']),allow_pickle=False)
|
||||
return np.asarray(x,np.float64),valid_rows(x,p['status'],p['nodes'])
|
||||
|
||||
|
||||
def moments(chunks):
|
||||
n=0;mean=np.zeros(2,np.float64);m2=np.zeros(2,np.float64)
|
||||
for x,valid in chunks:
|
||||
values=x[valid];k=len(values)
|
||||
if not k:continue
|
||||
avg=values.mean(0);delta=avg-mean
|
||||
m2+=np.square(values-avg).sum(0)+delta*delta*n*k/(n+k)
|
||||
mean+=delta*k/(n+k);n+=k
|
||||
if n==0:raise ValueError('No valid learning nodes; no invented normalization')
|
||||
std=np.sqrt(m2/n);scale=np.where(std==0,1.,std)
|
||||
if not np.isfinite(mean).all() or not np.isfinite(scale).all():raise ValueError('Context normalization overflow')
|
||||
return dict(mean=mean.tolist(),scale=scale.tolist(),population_std=std.tolist(),valid_nodes=n)
|
||||
|
||||
|
||||
def encode(extra,valid,stats):
|
||||
out=np.zeros((len(extra),3),np.float32)
|
||||
out[valid,:2]=((extra[valid]-np.asarray(stats['mean']))/np.asarray(stats['scale'])).astype(np.float32)
|
||||
out[valid,2]=1
|
||||
if not np.isfinite(out).all():raise ValueError('Context standardized overflow')
|
||||
return out
|
||||
|
||||
|
||||
def standardization(old,c,root):
|
||||
start=time.process_time();stats=original_standardization(old,root);seen=[]
|
||||
def chunks():
|
||||
for case in old['folds'][0]['learning']:
|
||||
x,valid=read_context(old,c,case);seen.append(dict(case=case,nodes=len(x),valid_nodes=int(valid.sum())))
|
||||
yield x,valid
|
||||
context=moments(chunks());context.update(parts=seen,weighting='unit supervised node as original H2',labels_read=False,cpu_seconds=time.process_time()-start)
|
||||
stats['context']=context;save(root/'standardization.json',stats);return stats
|
||||
|
||||
|
||||
def load_part(old,c,case,stats,role,root):
|
||||
# The original role guard runs before any extra context read.
|
||||
data=original_load(old,case,stats,role,root)
|
||||
extra,valid=read_context(old,c,case)
|
||||
data['x']=torch.cat([data['x'],torch.from_numpy(encode(extra,valid,stats['context']))],dim=1)
|
||||
data['context_valid_nodes']=int(valid.sum())
|
||||
return data
|
||||
@@ -0,0 +1,136 @@
|
||||
"""One item b attempt, scoped process memory plus failed-worker peak checks."""
|
||||
import ctypes
|
||||
import json
|
||||
import os
|
||||
from pathlib import Path
|
||||
import resource
|
||||
import signal
|
||||
import subprocess
|
||||
import sys
|
||||
import threading
|
||||
import time
|
||||
import psutil
|
||||
from phase2_h1_worker_v2 import save,sha
|
||||
from phase3_item_b_common import config,assignment,read
|
||||
from phase3_shared_capacity import validate_lease,alive
|
||||
from track_a_recovery_monitor import Monitor
|
||||
from track_a_recovery_supervisor import supervise,tree_usage,stop_group,disk_usage
|
||||
from s2_main_memory_guard import PeakRecords
|
||||
|
||||
|
||||
def run(queue,index,digest):
|
||||
parent=os.getppid()
|
||||
if ctypes.CDLL(None,use_errno=True).prctl(1,signal.SIGTERM,0,0,0)!=0:
|
||||
raise OSError('PDEATHSIG')
|
||||
requested=[]
|
||||
for sig in (signal.SIGTERM,signal.SIGINT):
|
||||
signal.signal(sig,lambda n,f:requested.append('SIGNAL_'+str(n)))
|
||||
if os.getppid()!=parent:
|
||||
raise RuntimeError('Queue lost before setup')
|
||||
c,_=config(queue);job=assignment(queue,c,index,digest)
|
||||
root=queue/'jobs'/str(index);start=time.monotonic();cpu=time.process_time()
|
||||
original=read(root/'assignment.json');wait=original['pre_dispatch_wait_seconds']
|
||||
remaining=c['runaway_gpu_seconds']-wait-60
|
||||
if remaining<90:
|
||||
raise RuntimeError('No cleanup window before worker')
|
||||
deadline=start+remaining
|
||||
# Queue writes ready only after atomically transferring reservation ownership.
|
||||
while not (root/'lease-ready.json').exists():
|
||||
if requested or not alive(job['queue_identity']) or time.monotonic()-start>30:
|
||||
raise RuntimeError('Lease transfer timeout')
|
||||
time.sleep(.1)
|
||||
validate_lease(job['token'],queue,index,sha(queue/'contract.json'))
|
||||
if os.getppid()!=job['queue_identity']['pid'] or not alive(job['queue_identity']):
|
||||
raise RuntimeError('Pinned queue lost before guard import completed')
|
||||
os.sched_setaffinity(0,[26,27])
|
||||
tracker=PeakRecords();done=threading.Event();monitor=Monitor(root/'monitor-errors.jsonl')
|
||||
reasons=[];receipt=None
|
||||
command=[sys.executable,'tools/phase3_item_b_worker.py',str(queue),str(index),digest]
|
||||
def reason(s):
|
||||
if s not in reasons:reasons.append(s)
|
||||
def workers():
|
||||
result=[]
|
||||
for p in psutil.Process().children(recursive=True):
|
||||
try:
|
||||
if p.cmdline()==command:result.append(p.pid)
|
||||
except psutil.NoSuchProcess:pass
|
||||
return result
|
||||
def kill():
|
||||
for pid in workers():stop_group(pid)
|
||||
def scan():
|
||||
tracker.scan(root/'worker')
|
||||
if max(tracker.maxima.values())>job['recorded_vram_bytes']:reason('RECORDED_VRAM_LIMIT')
|
||||
def local_watch():
|
||||
while not done.is_set():
|
||||
monitor.sample('recorded_peaks',scan)
|
||||
usage,ok=monitor.sample('tree',lambda:tree_usage(os.getpid()))
|
||||
if ok and (usage[0]>job['cpu_seconds']-30 or usage[1]>job['rss_bytes']):reason('GUARD_TREE_LIMIT')
|
||||
disk,ok=monitor.sample('disk',lambda:disk_usage(root))
|
||||
if ok and disk>job['disk_bytes']:reason('OUTPUT_LIMIT')
|
||||
if time.monotonic()>deadline-30:reason('OWN_ACTUAL_TIME_LIMIT')
|
||||
queue_live,ok=monitor.sample('queue_identity',lambda:alive(job['queue_identity']))
|
||||
if ok and not queue_live:reason('QUEUE_LOST')
|
||||
if (queue/'global-stop.json').exists():reason('GLOBAL_STOP')
|
||||
for s in requested:reason(s)
|
||||
if monitor.failed:reason('MONITOR_FAILED')
|
||||
if reasons:kill()
|
||||
done.wait(.5)
|
||||
def gpu_watch():
|
||||
def sample():
|
||||
text=subprocess.check_output(['nvidia-smi','--query-compute-apps=gpu_uuid,pid,used_gpu_memory','--format=csv,noheader,nounits'],text=True,timeout=10)
|
||||
rows=[]
|
||||
for line in text.strip().splitlines():
|
||||
if not line:continue
|
||||
u,p,m=[x.strip() for x in line.split(',')]
|
||||
if u==job['gpu_uuid']:rows.append(dict(pid=int(p),bytes=int(m)*2**20))
|
||||
if len(rows)>3:reason('HOST_GPU_SLOT_LIMIT')
|
||||
own=set(workers())
|
||||
if sum(r['bytes'] for r in rows if r['pid'] in own)>job['process_vram_bytes']:reason('OWN_GPU_PROCESS_MEMORY_LIMIT')
|
||||
save(root/'gpu-sample.json',dict(unix=time.time(),rows=rows,own_pids=sorted(own)))
|
||||
while not done.is_set():
|
||||
monitor.sample('gpu_processes',sample)
|
||||
if monitor.failed:reason('MONITOR_FAILED')
|
||||
if reasons:kill()
|
||||
done.wait(1)
|
||||
threads=[threading.Thread(target=f,daemon=True) for f in (local_watch,gpu_watch)]
|
||||
try:
|
||||
save(root/'started.json',dict(unix=time.time(),assignment_sha256=digest))
|
||||
for t in threads:t.start()
|
||||
if reasons or requested or time.monotonic()>deadline-60:raise RuntimeError('Prelaunch guard rejected')
|
||||
receipt=supervise(command,root/'guards','train',dict(total_cpu_seconds=job['cpu_seconds']-60,
|
||||
job_cpu_seconds=job['cpu_seconds']-60,job_wall_seconds=deadline-time.monotonic()-40,
|
||||
tree_rss_bytes=job['rss_bytes'],disk_bytes=job['disk_bytes'],max_processes=1),poll=.2,gpu_uuid=job['gpu_uuid'])
|
||||
except BaseException as exc:
|
||||
reason(type(exc).__name__+': '+str(exc));kill()
|
||||
finally:
|
||||
done.set()
|
||||
for t in threads:
|
||||
if t.ident is not None:t.join(timeout=12)
|
||||
if any(t.is_alive() for t in threads):reason('MONITOR_DID_NOT_EXIT')
|
||||
try:monitor.retry('final_peaks',scan,.1)
|
||||
except Exception as exc:reason('FINAL_PEAK_SCAN: '+str(exc))
|
||||
for s in requested:reason(s)
|
||||
if monitor.failed:reason('MONITOR_FAILED')
|
||||
p=root/'worker/result.json';result=read(p) if p.exists() else None
|
||||
if not receipt or receipt['status']!='PASS' or receipt['returncode']!=0:reason('WORKER_NOT_PASS')
|
||||
if not result or result.get('status')!='COMPLETE' or result.get('fold')!=index or result.get('completed_epochs')!=200 or result.get('optimizer_steps')!=job['target_steps']:reason('NO_COMPLETE_IDENTITY')
|
||||
if result:
|
||||
from phase2_h1_worker_v2 import checked
|
||||
checked(result['checkpoint'],result['checkpoint_sha256'])
|
||||
if len(tracker.witnesses)!=2:reason('MISSING_RECORDED_PEAKS')
|
||||
elapsed=time.monotonic()-start
|
||||
child=resource.getrusage(resource.RUSAGE_CHILDREN)
|
||||
charged_cpu=time.process_time()-cpu+child.ru_utime+child.ru_stime
|
||||
if wait+60+elapsed>c['runaway_gpu_seconds'] or charged_cpu>job['cpu_seconds']:reason('FINAL_RESOURCE_LIMIT')
|
||||
try:config(queue)
|
||||
except Exception as exc:reason('FINAL_PINS: '+str(exc))
|
||||
save(root/'complete.json',dict(status='PASS' if not reasons else 'STOPPED_OR_FAILED',reasons=reasons,
|
||||
worker=receipt,worker_result_sha256=sha(p) if p.exists() else None,
|
||||
cpu_seconds_charged=charged_cpu,pre_dispatch_wait_seconds=wait,guard_wall_seconds=elapsed,
|
||||
recorded_memory=dict(maxima=tracker.maxima,witnesses=tracker.witnesses,observations_count=len(tracker.observations)),
|
||||
monitor_policy=monitor.policy(),monitor_errors=monitor.records,no_retry=True))
|
||||
return 0 if not reasons else 1
|
||||
|
||||
|
||||
if __name__=='__main__':
|
||||
raise SystemExit(run(Path(sys.argv[1]),int(sys.argv[2]),sys.argv[3]))
|
||||
@@ -0,0 +1,18 @@
|
||||
"""Original H2 initialization plus three zero-weight input columns, +384 parameters."""
|
||||
import torch
|
||||
from torch import nn
|
||||
from phase2_h2_model import H2Geo
|
||||
|
||||
|
||||
class H2Context(H2Geo):
|
||||
def __init__(self):
|
||||
super().__init__()
|
||||
# No new random draws: shared parameters and subsequent RNG match H2.
|
||||
for layer in (self.encoder[0],self.head[0]):
|
||||
layer.weight=nn.Parameter(torch.cat([layer.weight.detach(),layer.weight.new_zeros((64,3))],dim=1))
|
||||
layer.in_features+=3
|
||||
|
||||
def forward(self,x,*args,**kwargs):
|
||||
if x.ndim!=2 or x.shape[1]!=22:
|
||||
raise ValueError('Exactly19 original plus2 context and1 validity columns')
|
||||
return super().forward(x,*args,**kwargs)
|
||||
@@ -0,0 +1,144 @@
|
||||
"""Review-gated five-fold b training. Stops at frozen models, never postfit."""
|
||||
import fcntl
|
||||
import json
|
||||
import os
|
||||
from pathlib import Path
|
||||
import signal
|
||||
import subprocess
|
||||
import sys
|
||||
import time
|
||||
import psutil
|
||||
from phase2_h1_worker_v2 import save,sha
|
||||
from phase2_home_freeze_io import freeze
|
||||
from phase2_gate_outcome_review import validate_review
|
||||
from phase3_item_b_common import ROOT,config,read,project
|
||||
from phase3_item_b_capacity import register_entry,acquire,transfer,release,process_identity,alive
|
||||
from track_a_recovery_monitor import Monitor
|
||||
|
||||
|
||||
def run(root):
|
||||
c,_=config(root);validate_review(root)
|
||||
with (root/'queue.lock').open('a') as lock:
|
||||
fcntl.flock(lock,fcntl.LOCK_EX|fcntl.LOCK_NB)
|
||||
if (root/'started.json').exists():raise FileExistsError('Sole b attempt, no resume')
|
||||
if os.environ.get('PYTORCH_CUDA_ALLOC_CONF')!='expandable_segments:True':
|
||||
raise ValueError('Explicit allocator required')
|
||||
if psutil.disk_usage(root).free<100*2**30:raise RuntimeError('Less than100GiB free')
|
||||
register_entry(root,sha(root/'contract.json'))
|
||||
start=time.monotonic();identity=process_identity(os.getpid())
|
||||
save(root/'started.json',dict(unix=time.time(),identity=identity,contract_sha256=sha(root/'contract.json')))
|
||||
(root/'jobs').mkdir(exist_ok=False)
|
||||
active={};terminal={};pending=[0,1,2,3,4];histories={};stops=[];failure=None
|
||||
monitor=Monitor(root/'admission-errors.jsonl')
|
||||
for sig in (signal.SIGTERM,signal.SIGINT):signal.signal(sig,lambda n,f:stops.append('SIGNAL_'+str(n)))
|
||||
def charge(i):
|
||||
return terminal[i]['gpu_seconds_charged'] if i in terminal else 60+time.monotonic()-start
|
||||
def stop(reason):
|
||||
if not (root/'global-stop.json').exists():save(root/'global-stop.json',dict(reason=reason,unix=time.time()))
|
||||
try:
|
||||
while active or pending:
|
||||
if stops:raise RuntimeError(stops[0])
|
||||
if (root/'global-stop.json').exists():raise RuntimeError('EXTERNAL_GLOBAL_STOP')
|
||||
# Collect exits before granting any further admission.
|
||||
for i,(p,lease,log,owner) in list(active.items()):
|
||||
if p.poll() is None:continue
|
||||
log.close();jr=root/'jobs'/str(i);cp=jr/'complete.json'
|
||||
result=read(cp) if cp.exists() else dict(status='MISSING_GUARD_RECEIPT')
|
||||
worker_identity=jr/'worker/worker-identity.json'
|
||||
if worker_identity.exists() and alive(read(worker_identity)):
|
||||
raise RuntimeError('WORKER_OUTLIVED_GUARD_LEASE_HELD')
|
||||
terminal[i]=dict(id=i,status=result['status'],returncode=p.returncode,
|
||||
gpu_seconds_charged=charge(i),guard_sha256=sha(cp) if cp.exists() else None)
|
||||
release(lease['token'],owner);del active[i]
|
||||
if result['status']!='PASS' or p.returncode!=0:raise RuntimeError('FOLD_FAILED_'+str(i))
|
||||
for i in active:
|
||||
folder=root/'jobs'/str(i)/'worker/history'
|
||||
histories[i]=[read(p) for p in sorted(folder.glob('*.json'))]
|
||||
charges={i:charge(i) for i in range(5)}
|
||||
projection=project(c,charges,histories,terminal)
|
||||
save(root/'latest-cost-projection.json',projection)
|
||||
if projection['status']!='PASS':raise RuntimeError('TOTAL_PROJECTION_OR_ACTUAL_CAP')
|
||||
for i in active:
|
||||
gate=root/'jobs'/str(i)/'first-epoch-projection.json'
|
||||
if histories.get(i) and not gate.exists():
|
||||
save(gate,dict(status='PASS',contract_sha256=sha(root/'contract.json'),projection=projection))
|
||||
save(root/'live-progress.json',dict(unix=time.time(),stage='ITEM_B_TRAINING',active=list(active),pending=pending,
|
||||
terminal=list(terminal.values()),charged_training_gpu_seconds=projection['actual_training_and_preparation_gpu_seconds'],postfit_started=False))
|
||||
if pending:
|
||||
i=pending[0]
|
||||
decision,ok=monitor.sample('admission',lambda:acquire(root,i,sha(root/'contract.json')))
|
||||
if monitor.failed:raise RuntimeError('THREE_ADMISSION_FAILURES')
|
||||
if not ok:
|
||||
time.sleep(1)
|
||||
continue
|
||||
lease,observation=decision
|
||||
save(root/'latest-admission.json',observation)
|
||||
# Acquisition includes bounded GPU queries; recheck cost before Popen.
|
||||
fresh=project(c,{j:charge(j) for j in range(5)},histories,terminal)
|
||||
if fresh['status']!='PASS':
|
||||
if lease:release(lease['token'])
|
||||
raise RuntimeError('CAP_DURING_ADMISSION')
|
||||
if lease:
|
||||
jr=root/'jobs'/str(i);jr.mkdir(exist_ok=False)
|
||||
save(jr/'assignment.json',dict(id=i,contract_sha256=sha(root/'contract.json'),
|
||||
gpu_uuid=lease['gpu_uuid'],worker_cpus=lease['worker_cpus'],token=lease['token'],
|
||||
queue_identity=identity,pre_dispatch_wait_seconds=time.monotonic()-start))
|
||||
log=(jr/'guard.log').open('x')
|
||||
env=dict(os.environ,OPENBLAS_NUM_THREADS='1',OMP_NUM_THREADS='1',MKL_NUM_THREADS='1')
|
||||
p=subprocess.Popen([sys.executable,'tools/phase3_item_b_guard.py',str(root),str(i),sha(jr/'assignment.json')],
|
||||
cwd=str(Path(__file__).resolve().parents[1]),env=env,stdout=log,stderr=subprocess.STDOUT,start_new_session=True)
|
||||
owner=process_identity(p.pid)
|
||||
active[i]=(p,lease,log,owner);pending.remove(i)
|
||||
transfer(lease['token'],owner)
|
||||
save(jr/'lease-ready.json',dict(token=lease['token'],owner=owner))
|
||||
time.sleep(1)
|
||||
except BaseException as exc:
|
||||
failure=type(exc).__name__+': '+str(exc);stop(failure)
|
||||
finally:
|
||||
if active:
|
||||
stop(failure or 'QUEUE_CLEANUP')
|
||||
for p,lease,log,owner in active.values():
|
||||
if p.poll() is None:p.terminate()
|
||||
deadline=time.monotonic()+45
|
||||
while time.monotonic()<deadline and any(p.poll() is None for p,_,_,_ in active.values()):time.sleep(.2)
|
||||
for i,(p,lease,log,owner) in active.items():
|
||||
if p.poll() is None:p.kill()
|
||||
p.wait();log.close();jr=root/'jobs'/str(i);cp=jr/'complete.json'
|
||||
result=read(cp) if cp.exists() else dict(status='MISSING_GUARD_RECEIPT')
|
||||
terminal[i]=dict(id=i,status=result['status'],returncode=p.returncode,gpu_seconds_charged=charge(i),
|
||||
guard_sha256=sha(cp) if cp.exists() else None)
|
||||
wp=jr/'worker/worker-identity.json'
|
||||
try:
|
||||
if wp.exists() and alive(read(wp)):raise RuntimeError('Live worker; lease retained')
|
||||
release(lease['token'],owner)
|
||||
except Exception as exc:
|
||||
terminal[i]['lease_release_failure']=str(exc);failure=failure or str(exc)
|
||||
final_charges={i:charge(i) for i in range(5)}
|
||||
final_cost=c['preparation_charge_gpu_seconds']+sum(final_charges.values())
|
||||
if final_cost+c['postfit_gpu_reserve']>c['gpu_cap']:failure=failure or 'FINAL_CUMULATIVE_CAP'
|
||||
complete=not failure and len(terminal)==5 and all(r['status']=='PASS' and r['returncode']==0 for r in terminal.values())
|
||||
files=[root/'contract.json']
|
||||
if complete:
|
||||
models=[]
|
||||
for i in range(5):
|
||||
r=read(root/'jobs'/str(i)/'worker/result.json')
|
||||
models.append(dict(r));files.append(Path(r['checkpoint']))
|
||||
files.extend([root/'jobs'/str(i)/'complete.json',root/'jobs'/str(i)/'worker/result.json',root/'jobs'/str(i)/'worker/standardization.json'])
|
||||
save(root/'models.json',models);files.append(root/'models.json')
|
||||
save(root/'complete.json',dict(status='TRAIN_COMPLETE_POSTFIT_PENDING' if complete else 'STOPPED_INCOMPLETE',
|
||||
failure=failure,entries=list(terminal.values()),never_attempted=pending,
|
||||
never_attempted_wait_charges={str(i):final_charges[i] for i in pending},
|
||||
cumulative_gpu_seconds=final_cost,postfit_gpu_reserve=c['postfit_gpu_reserve'],
|
||||
no_retry=True,test_targets_read=False,prediction_started=False,scoring_started=False))
|
||||
files.append(root/'complete.json')
|
||||
freeze_start=time.monotonic()
|
||||
freeze(root,'training-terminal' if complete else 'training-stop',files)
|
||||
final_cost+=time.monotonic()-freeze_start
|
||||
within=final_cost+c['postfit_gpu_reserve']<=c['gpu_cap']
|
||||
save(root/'post-terminal-cost.json',dict(cumulative_gpu_seconds=final_cost,within_cap=within,
|
||||
freeze_tail_charged_once=True,postfit_gpu_reserve=c['postfit_gpu_reserve'],postfit_started=False,
|
||||
training_complete_sha256=sha(root/'complete.json'),required_before_any_postfit=True))
|
||||
return 0 if complete and within else 1
|
||||
|
||||
|
||||
if __name__=='__main__':raise SystemExit(run(ROOT))
|
||||
@@ -0,0 +1,79 @@
|
||||
"""One new S2 fold, fixed original200epoch schedule with seed0; no test prediction or target reads."""
|
||||
import json,os,sys,time,traceback,random
|
||||
from pathlib import Path
|
||||
import numpy as np
|
||||
import torch
|
||||
from phase2_h2_s1_common import selected_checkpoint
|
||||
from phase2_h2_model import lr_epoch
|
||||
from phase3_item_b_common import config,fold_view,first_epoch_release,assignment as assigned_job,read
|
||||
from phase3_shared_capacity import process_identity
|
||||
from phase3_item_b_data import standardization,load_part
|
||||
from phase2_h2_data import to_device,arguments,cached
|
||||
from phase3_item_b_model import H2Context as H2Geo
|
||||
from phase2_h2_model import optimizer
|
||||
from phase2_h2_pilot_worker import peaks,train_step
|
||||
from phase2_residual_math import residual_target
|
||||
from phase2_h1_worker_v2 import save,sha
|
||||
|
||||
def run(root,fold_id,assignment_sha):
|
||||
c,pilot=config(root);old=fold_view(pilot,fold_id);job=assigned_job(root,c,fold_id,assignment_sha);out=root/'jobs'/str(fold_id)/'worker';out.mkdir(exist_ok=False);(out/'history').mkdir()
|
||||
os.sched_setaffinity(0,job['worker_cpus']);torch.set_num_threads(1);torch.set_num_interop_threads(1)
|
||||
start=time.monotonic();cpu=time.process_time();stage='SETUP';epoch=0;steps=0
|
||||
save(out/'worker-identity.json',process_identity(os.getpid()))
|
||||
def record(item,epoch_file=False):
|
||||
item=dict(fold=fold_id,stage=stage,epoch=epoch,optimizer_steps=steps,wall_seconds=time.monotonic()-start,cpu_seconds=time.process_time()-cpu,**item,**peaks())
|
||||
with (out/'history.jsonl').open('a') as f:f.write(json.dumps(item,allow_nan=False)+'\n')
|
||||
save(out/'progress.json',item)
|
||||
if epoch_file:save(out/'history'/f'{epoch:03d}.json',item)
|
||||
return item
|
||||
try:
|
||||
if os.environ.get('CUDA_VISIBLE_DEVICES')!=job['gpu_uuid'] or os.environ.get('PYTORCH_CUDA_ALLOC_CONF')!='expandable_segments:True':raise ValueError('GPU/allocator identity')
|
||||
stage='LEARNING_ONLY_NORMALIZATION';normal=out/'normalization';normal.mkdir();stats=standardization(old,c,normal);save(out/'standardization.json',stats);save(out/'fold-identity.json',dict(actual_fold=fold_id,learning=old['folds'][0]['learning'],validation=old['folds'][0]['validation'],test=old['folds'][0]['test'],clock_sha256=old['folds'][0]['clock_sha256']))
|
||||
torch.backends.cuda.matmul.allow_tf32=False;torch.backends.cudnn.allow_tf32=False
|
||||
torch.cuda.set_device(0);torch.cuda.init();torch.cuda.reset_peak_memory_stats();random.seed(0);np.random.seed(0);torch.manual_seed(0);torch.cuda.manual_seed_all(0)
|
||||
model=H2Geo().cuda();opt=optimizer(model);fold=old['folds'][0];parts=len(fold['learning']);total=fold['learning_nodes']
|
||||
save(out/'environment.json',dict(torch=torch.__version__,gpu_uuid=job['gpu_uuid'],allocator_settings=str(torch.cuda.memory._snapshot().get('allocator_settings',{})),model_dtype='FP32',autocast=False,grad_scaler=False,fresh_seed=0,pilot_weights_reused=False,parameter_count=sum(p.numel() for p in model.parameters())))
|
||||
for epoch in range(1,201):
|
||||
stage='TRAIN';begin=time.monotonic();seen=0;nodes=0;loss_sum=0.
|
||||
if any(abs(g['lr']-lr_epoch(epoch))>1e-15 for g in opt.param_groups):raise ValueError('Learning rate epoch mismatch')
|
||||
order=np.random.default_rng(np.random.SeedSequence([0,epoch])).permutation(parts)
|
||||
for index in order:
|
||||
case=fold['learning'][int(index)];data=to_device(load_part(old,c,case,stats,'learning',out),'cuda')
|
||||
result=train_step(model,opt,data,parts,total);seen+=1;steps+=1;nodes+=data['nodes'];loss_sum+=result['loss']
|
||||
record(dict(case=case,epoch_parts=seen,epoch_nodes=nodes,lr=lr_epoch(epoch),loss=result['loss'],gradient_norm_before_clip=result['gradient_norm_before_clip']))
|
||||
del data
|
||||
if seen!=parts or nodes!=total:raise ValueError('Incomplete full-node learning epoch')
|
||||
metrics=dict(train_mean_weighted_loss=loss_sum/parts,train_epoch_seconds=time.monotonic()-begin,epoch_parts=seen,epoch_nodes=nodes,lr=lr_epoch(epoch))
|
||||
if epoch%10==0:
|
||||
stage='VALIDATION_DIAGNOSTIC';model.eval();val_loss=0.;val_nodes=0;vstart=time.monotonic()
|
||||
with torch.no_grad():
|
||||
for case in fold['validation']:
|
||||
data=to_device(load_part(old,c,case,stats,'validation',out),'cuda');p=model(*arguments(data),checkpoint_blocks=False)
|
||||
if not bool(torch.isfinite(p).all()):raise ValueError('Nonfinite validation output')
|
||||
with (out/'validation-target-access.jsonl').open('a') as f:f.write(json.dumps(dict(case=case,epoch=epoch,role='validation_diagnostic_only',target_sha256=old['parts'][case]['output_sha256']['target.npy']))+'\n')
|
||||
target=residual_target(cached(old,case,'target'),data['F'],fold['c'],fit_dtype='float32')
|
||||
delta=p.cpu().numpy().astype(np.float64)-target;val_loss+=float(delta@delta);val_nodes+=data['nodes'];del data,p
|
||||
torch.cuda.synchronize()
|
||||
if val_nodes!=fold['validation_nodes']:raise ValueError('Validation coverage')
|
||||
metrics.update(validation_mse=val_loss/val_nodes,validation_nodes=val_nodes,validation_seconds=time.monotonic()-vstart,validation_used_for_selection=False)
|
||||
stage='EPOCH_COMPLETE';record(metrics,epoch_file=True)
|
||||
if epoch==1:
|
||||
stage='WAIT_THIS_FOLD_FIRST_EPOCH_COST_GATE';record(dict(epoch_parts=seen,epoch_nodes=nodes))
|
||||
while not first_epoch_release(root,fold_id):time.sleep(.5)
|
||||
if epoch<200:
|
||||
for group in opt.param_groups:group['lr']=lr_epoch(epoch+1)
|
||||
stage='FINAL_CHECKPOINT';name=selected_checkpoint(epoch)
|
||||
torch.save(dict(model=model.state_dict(),optimizer=opt.state_dict(),completed_epochs=epoch,optimizer_steps=steps,seed=0,fold=fold_id,contract_sha256=sha(root/'contract.json'),role='last_epoch_only'),out/name)
|
||||
save(out/'result.json',dict(status='COMPLETE',fold=fold_id,completed_epochs=epoch,optimizer_steps=steps,learning_parts=parts,learning_nodes_per_epoch=total,checkpoint=str(out/name),checkpoint_sha256=sha(out/name),standardization_sha256=sha(out/'standardization.json'),test_targets_read=False,test_prediction_started=False,validation_used_for_selection=False,worker_wall_seconds=time.monotonic()-start,worker_cpu_seconds=time.process_time()-cpu,**peaks()))
|
||||
except BaseException as exc:
|
||||
memory=peaks() if torch.cuda.is_initialized() else dict(peak_allocated_bytes=0,peak_reserved_bytes=0)
|
||||
save(out/'failure.json',dict(status='FAILED',stage=stage,epoch=epoch,optimizer_steps=steps,type=type(exc).__name__,reason=str(exc),traceback=traceback.format_exc(),worker_wall_seconds=time.monotonic()-start,worker_cpu_seconds=time.process_time()-cpu,**memory));raise
|
||||
if __name__=='__main__':
|
||||
import ctypes,signal
|
||||
from phase3_shared_capacity import alive
|
||||
expected=read(Path(sys.argv[1])/'jobs'/sys.argv[2]/'lease-ready.json')['owner']
|
||||
parent=os.getppid()
|
||||
if parent!=expected['pid'] or not alive(expected):raise RuntimeError('Pinned guard lost before worker import completed')
|
||||
if ctypes.CDLL(None,use_errno=True).prctl(1,signal.SIGKILL,0,0,0)!=0:raise OSError('PDEATHSIG')
|
||||
if os.getppid()!=parent:raise RuntimeError('Guard exited before worker setup')
|
||||
run(Path(sys.argv[1]),int(sys.argv[2]),sys.argv[3])
|
||||
@@ -0,0 +1,79 @@
|
||||
"""Metadata and opaque hashes only; original552 B13 cache, no new geometry/labels/fits."""
|
||||
import time
|
||||
from pathlib import Path
|
||||
from phase2_h1_worker_v2 import checked,sha,save
|
||||
from phase3_item_b_common import ROOT,read
|
||||
from phase2_gate_outcome_review import validate_review
|
||||
from track_a_h2_prepare import local_code_closure
|
||||
from phase3_shared_capacity import PHASE_RESERVES
|
||||
|
||||
B=Path(__file__).resolve().parents[1]
|
||||
H=Path('/home/frank14f/MeltMind-runs')
|
||||
|
||||
|
||||
def main():
|
||||
start=time.process_time()
|
||||
if ROOT.exists():raise FileExistsError('Sole item b root already exists')
|
||||
prior=H/'phase3-item-1b-2026-11-18';validate_review(prior)
|
||||
pc=read(prior/'contract.json');pilot=read(pc['pilot_contract'])
|
||||
spec=B/'references/2026-11-18/phase3-item-b/mold-regional-context-spec.md'
|
||||
checked(spec,'de120e9dc5527c01af5247a27c8cf529985c206821553bacb01be6071f2cdab5')
|
||||
old=B/'artifacts/phase2-b13-e1b-2026-11-17';oc=read(old/'contract.json')
|
||||
if read(old/'complete.json')['status']!='COMPLETE' or oc['new_geometry_indices'] or len(oc['reused_geometry'])!=552:
|
||||
raise ValueError('Original complete552 context reuse required')
|
||||
source=dict(pc['source_sha256'])
|
||||
for p in [spec,prior/'contract.json',prior/'review.json',prior/'expected-review-binding.json',prior/'implementation-review-git-freeze.json',prior/'implementation-review-authorization.json',old/'contract.json',old/'features-manifest.json',old/'complete.json']:
|
||||
source[str(p)]=sha(p)
|
||||
entries={r['case']:r for r in read(old/'features-manifest.json')['entries']}
|
||||
reused={r['case']:r for r in oc['reused_geometry']};context={}
|
||||
roster={r['case']:r for r in read(pc['query_roster_contract'])['rows']}
|
||||
if set(entries)!=set(roster) or len(entries)!=552:raise ValueError('Original full roster differs')
|
||||
for row in oc['parts']:
|
||||
case=row['case'];r=pilot['parts'][case];e=entries[case];reuse=reused[case]
|
||||
if row['nodes']!=r['nodes'] or row['component_id']!=roster[case]['component_id'] or row['v5_cache']['output_sha256']!=r['output_sha256'] or row['preparation']!=roster[case]['preparation']:
|
||||
raise ValueError('Frozen context node/geometry/cache/query identity')
|
||||
p=old/'parts'/f"{e['index']:04d}";result=read(checked(p/'result.json',e['result_sha256']))
|
||||
if reuse['index']!=e['index'] or reuse['result_sha256']!=e['result_sha256'] or reuse['extra_sha256']!=result['extra_sha256'] or result['case']!=case:
|
||||
raise ValueError('Original E1b immutable reused entry differs')
|
||||
if result['status'] not in ('GEOMETRY_COMPLETE','SCIENTIFIC_MISSING') or result['kind']!='B13' or result['labels_read']:
|
||||
raise ValueError('Only actual scientific missingness, no engineering substitution')
|
||||
extra=checked(p/'extra.npy',result['extra_sha256'])
|
||||
for q in (p/'result.json',extra):source[str(q)]=sha(q)
|
||||
context[case]=dict(case=case,nodes=r['nodes'],status=result['status'],result=str(p/'result.json'),result_sha256=sha(p/'result.json'),extra=str(extra),extra_sha256=sha(extra),raw_ids_sha256=r['output_sha256']['raw_ids.npy'],features_sha256=r['output_sha256']['features.npy'],component_id=row['component_id'])
|
||||
jobs=[dict(j,target_steps=200*len(j['learning']),cpu_seconds=60000) for j in pc['jobs']]
|
||||
files=sorted(B.glob('tools/phase3_item_b_*.py'))+[Path(__file__).resolve(),B/'tools/test_phase3_item_b.py']
|
||||
codes=local_code_closure(files+sorted((B/'tools/s2_models').glob('*.py')))
|
||||
for p,h in pc['code_sha256'].items():codes.setdefault(p,h)
|
||||
for p,h in {**source,**codes}.items():checked(p,h)
|
||||
comparison=dict(version='item-b-training-v1',spec_sha256=sha(spec),parallel_ruling='1795006714.455999',budget_ruling='1795007580.895539',entry_ruling='1795011607.907609',
|
||||
folds=pilot['folds'],roster_parts=552,components=204,seed=0,epochs=200,inputs=22,added_parameters=384,
|
||||
intervention='original H2 plus frozen two B13 context means and joint valid flag; full roster retained',
|
||||
initialization='original shared seed0 weights and RNG preserved; encoder19->22/head83->86 added weights zero',
|
||||
normalization='original19 unchanged; context mean/population scale from valid learning nodes only, unit node weighting, zero variance scale1; invalid standardized context0 flag0, valid flag1',
|
||||
missingness='whole-part joint NaN only with historical SCIENTIFIC_MISSING; asymmetric/nonfinite/ID/cache errors stop',
|
||||
geometry='all552 frozen B13 readouts reused by original result/extra/node/feature/geometry identity; no new METIS/partition/K/seed',
|
||||
recipe='same cF FP64 residual and final200epoch schedule/seed0/MSE/AdamW/clipping/part shuffle; diagnostic validation every10 only',
|
||||
baseline_models=pc['baseline_models'],seed1_root=str(H/'phase3-seed1-v2-2026-11-18'),
|
||||
test=dict(requires_separate_pre_score_review=True,requires_seed_reference_for_decision=True,parts=552,components=204,
|
||||
upper_delta_relativeL2_strictly_below=-.001,absolute_effect_strictly_above_absolute_pooled_seed_difference=True,
|
||||
bootstrap=2000,bootstrap_seed=2026111408,paired_components=True,confidence=.95,
|
||||
report=['signed means and CI','seed difference same endpoint','G valid/invalid parts and components','G subgroup paired differences','per-fold','hard last-point/Rg','tau','negative times','end-to-end runtime including uncached partition cost as unmeasured']),
|
||||
postfit='only after five terminal models frozen and all workers exit; own reviewed implementation, no automatic postfit in queue',
|
||||
limitations=['increment of two context columns and missingness together','benefit on invalid subset not spatial attribution','restricted development-exposed corpus not fresh family proof','not flow conservation','does not close independent type/connectivity/O1 hypotheses'])
|
||||
ROOT.mkdir();save(ROOT/'comparison-contract.json',comparison)
|
||||
c=dict(version='phase3-item-b-v1',approval='1795011607.907609',parallel_approval='1795006714.455999',budget_approval='1795007580.895539',
|
||||
seed=0,epochs=200,new_fits=5,inputs=22,added_parameters=384,no_retry=True,gpu_cap=178200,postfit_gpu_reserve=7200,postfit_cpu_cap=3600,
|
||||
phase3_gpu_cap=720000,phase_reservations=PHASE_RESERVES,preparation_charge_gpu_seconds=600,preparation_cpu_cap=600,
|
||||
runaway_gpu_seconds=60000,decision_requires_seed_reference=True,dispatch_order=[0,1,2,3,4],
|
||||
waiting_charge='all5 from queue start to own terminal; includes pending never-dispatched waits/failures;60s setup per fold',
|
||||
projection='actual+1.5 remaining200epoch training and20 diagnostic validations+120 cleanup; total gate, no per-fold projection stop',
|
||||
validation_targets_read='diagnostic only every10epochs after reviewed launch',test_targets_read=False,postfit_started=False,track_a_ledger='separate',
|
||||
shared_rule_module='phase3_shared_capacity.py unchanged reviewed2ddecef3/e30612ea',capacity_adapter='b key/cap and GPU0..2 only; shared admission/registry/leases unchanged',
|
||||
jobs=jobs,pilot_contract=pc['pilot_contract'],query_roster_contract=pc['query_roster_contract'],baseline_models=pc['baseline_models'],context=context,
|
||||
source_sha256=source,code_sha256=codes,comparison_contract=str(ROOT/'comparison-contract.json'),comparison_contract_sha256=sha(ROOT/'comparison-contract.json'))
|
||||
save(ROOT/'contract.json',c)
|
||||
save(ROOT/'preparation-audit.json',dict(status='PREPARED_PENDING_ASSIST_REVIEW',contract_sha256=sha(ROOT/'contract.json'),code_files=len(codes),source_files=len(source),context_parts=len(context),valid_parts=sum(x['status']=='GEOMETRY_COMPLETE' for x in context.values()),scientific_missing_parts=sum(x['status']=='SCIENTIFIC_MISSING' for x in context.values()),opaque_cache_hashes_only=True,metadata_cpu_seconds=time.process_time()-start,labels_decoded=False,arrays_decoded=False,models_loaded=False,new_fits_started=False))
|
||||
print(read(ROOT/'preparation-audit.json'))
|
||||
|
||||
|
||||
if __name__=='__main__':main()
|
||||
@@ -0,0 +1,196 @@
|
||||
"""Synthetic b acceptance; no real data/model loading, GPU query or scientific run."""
|
||||
import ast,copy,inspect,json,tempfile,unittest
|
||||
from contextlib import ExitStack,contextmanager
|
||||
from pathlib import Path
|
||||
from types import SimpleNamespace
|
||||
from unittest.mock import patch
|
||||
import numpy as np
|
||||
import torch
|
||||
from phase3_item_b_common import project
|
||||
from phase3_shared_capacity import GPUS,GIB
|
||||
from phase3_item_b_data import valid_rows,moments,encode
|
||||
from phase3_item_b_model import H2Context
|
||||
from phase2_h2_model import H2Geo
|
||||
|
||||
class ContextTests(unittest.TestCase):
|
||||
def test_joint_missing_is_zero_after_normalization(self):
|
||||
x=np.full((3,2),np.nan);v=valid_rows(x,'SCIENTIFIC_MISSING',3)
|
||||
self.assertTrue(np.array_equal(encode(x,v,dict(mean=[7,8],scale=[1,2])),np.zeros((3,3))))
|
||||
def test_valid_mean_still_has_flag(self):
|
||||
x=np.array([[7.,8.]])
|
||||
self.assertTrue(np.array_equal(encode(x,valid_rows(x,'GEOMETRY_COMPLETE',1),dict(mean=[7,8],scale=[1,2])),[[0,0,1]]))
|
||||
def test_partial_missing_stops(self):
|
||||
for x in (np.array([[1.,np.nan]]),np.array([[np.nan,1.]])):
|
||||
with self.assertRaises(ValueError):valid_rows(x,'SCIENTIFIC_MISSING',1)
|
||||
def test_status_mismatch_not_scientific_missing(self):
|
||||
for x,status in ((np.zeros((2,2)),'SCIENTIFIC_MISSING'),(np.full((2,2),np.nan),'GEOMETRY_COMPLETE'),(np.array([[1.,2.],[np.nan,np.nan]]),'GEOMETRY_COMPLETE')):
|
||||
with self.assertRaises(ValueError):valid_rows(x,status,2)
|
||||
def test_infinity_or_wrong_shape_stops(self):
|
||||
for x in (np.array([[np.inf,np.inf]]),np.zeros((1,3))):
|
||||
with self.assertRaises(ValueError):valid_rows(x,'GEOMETRY_COMPLETE',1)
|
||||
def test_unit_node_moments_not_equal_part_means(self):
|
||||
a=np.array([[1.,2.],[3.,4.],[5.,8.]]);b=np.array([[9.,12.]])
|
||||
stats=moments([(a,np.ones(3,bool)),(b,np.ones(1,bool)),(np.full((2,2),np.nan),np.zeros(2,bool))])
|
||||
x=np.concatenate([a,b]);np.testing.assert_allclose(stats['mean'],x.mean(0));np.testing.assert_allclose(stats['scale'],x.std(0));self.assertEqual(stats['valid_nodes'],4)
|
||||
def test_zero_variance_scale_one(self):
|
||||
stats=moments([(np.full((4,2),7.),np.ones(4,bool))]);self.assertEqual(stats['scale'],[1.,1.])
|
||||
def test_all_learning_invalid_stops(self):
|
||||
with self.assertRaises(ValueError):moments([(np.full((3,2),np.nan),np.zeros(3,bool))])
|
||||
def test_learning_only_iteration(self):
|
||||
import phase3_item_b_data as d
|
||||
old=dict(folds=[dict(learning=['a','b'],validation=['v'],test=['t'])]);seen=[]
|
||||
with tempfile.TemporaryDirectory() as tmp,patch.object(d,'original_standardization',return_value=dict(mean=[0]*19,scale=[1]*19)),patch.object(d,'read_context',side_effect=lambda o,c,k:(seen.append(k) or np.ones((2,2)),np.ones(2,bool))):
|
||||
stats=d.standardization(old,{},Path(tmp))
|
||||
self.assertEqual(seen,['a','b']);self.assertEqual(stats['context']['valid_nodes'],4)
|
||||
def test_original_role_guard_precedes_context_read(self):
|
||||
import phase3_item_b_data as d
|
||||
with patch.object(d,'original_load',side_effect=PermissionError),patch.object(d,'read_context') as extra:
|
||||
with self.assertRaises(PermissionError):d.load_part({}, {},'t',{},'test',Path('/synthetic'))
|
||||
extra.assert_not_called()
|
||||
|
||||
|
||||
class ModelTests(unittest.TestCase):
|
||||
def pair(self):
|
||||
torch.manual_seed(0);old=H2Geo();rng=torch.get_rng_state().clone()
|
||||
torch.manual_seed(0);new=H2Context();self.assertTrue(torch.equal(rng,torch.get_rng_state()))
|
||||
return old,new
|
||||
def test_all_shared_parameters_and_rng_match(self):
|
||||
old,new=self.pair()
|
||||
for key,value in old.state_dict().items():
|
||||
v=new.state_dict()[key]
|
||||
if key in ('encoder.0.weight','head.0.weight'):
|
||||
self.assertTrue(torch.equal(value,v[:,:value.shape[1]]));self.assertEqual(int(torch.count_nonzero(v[:,value.shape[1]:])),0)
|
||||
else:self.assertTrue(torch.equal(value,v),key)
|
||||
def test_exactly384_additional_parameters(self):
|
||||
old,new=self.pair();self.assertEqual(sum(x.numel() for x in new.parameters())-sum(x.numel() for x in old.parameters()),384)
|
||||
def inputs(self):
|
||||
n=4;x=torch.randn(n,19);mass=torch.ones(n);r=torch.tensor([0,1,2,3]);s=torch.tensor([1,2,3,0]);a=torch.randn(n,4);ids=torch.zeros(n,dtype=torch.long)
|
||||
return x,(mass,r,s,a,ids)
|
||||
def test_nonzero_readout_still_matches_original_initial_function(self):
|
||||
old,new=self.pair();x,args=self.inputs()
|
||||
with torch.no_grad():old.head[-1].weight.fill_(.01);new.head[-1].weight.copy_(old.head[-1].weight)
|
||||
old.eval();new.eval()
|
||||
with torch.no_grad():
|
||||
a=old(x,*args);b=new(torch.cat([x,torch.randn(4,3)],dim=1),*args)
|
||||
torch.testing.assert_close(a,b,rtol=1e-5,atol=1e-6)
|
||||
def test_new_columns_can_learn_without_changing_recipe(self):
|
||||
from phase2_h2_model import optimizer
|
||||
from phase2_h2_pilot_worker import train_step
|
||||
_,new=self.pair();x,args=self.inputs();data=dict(zip(['mass','receiver','sender','attributes','component_ids'],args));data.update(x=torch.cat([x,torch.ones(4,3)],dim=1),target=torch.ones(4),nodes=4)
|
||||
opt=optimizer(new)
|
||||
# Only synchronize is replaced; real original FP32 train_step/backward is exercised on CPU.
|
||||
with patch.object(torch.cuda,'synchronize',lambda:None):
|
||||
train_step(new,opt,data,1,4);train_step(new,opt,data,1,4)
|
||||
self.assertGreater(float(new.head[0].weight[:,-3:].abs().sum()),0)
|
||||
def test_wrong_input_columns_rejected(self):
|
||||
_,new=self.pair();x,args=self.inputs()
|
||||
with self.assertRaises(ValueError):new(x,*args)
|
||||
def test_worker_uses_original200_recipe(self):
|
||||
s=(Path(__file__).parent/'phase3_item_b_worker.py').read_text()
|
||||
self.assertIn('range(1,201)',s);self.assertIn('SeedSequence([0,epoch])',s);self.assertIn('if epoch%10==0:',s)
|
||||
self.assertIn('from phase2_h2_model import lr_epoch',s);self.assertIn('from phase2_h2_pilot_worker import peaks,train_step',s)
|
||||
self.assertNotIn('location_loss',s);self.assertNotIn('screen_decision',s)
|
||||
def test_guard_accepts_only_complete200_epoch_identity(self):
|
||||
s=(Path(__file__).parent/'phase3_item_b_guard.py').read_text()
|
||||
self.assertIn("result.get('completed_epochs')!=200",s)
|
||||
self.assertNotIn("result.get('completed_epochs')!=100",s)
|
||||
|
||||
class BudgetTests(unittest.TestCase):
|
||||
def setUp(self):self.c=dict(gpu_cap=178200,postfit_gpu_reserve=7200,preparation_charge_gpu_seconds=600,runaway_gpu_seconds=60000,jobs=[dict(validation_seconds_bound=3) for _ in range(5)])
|
||||
def test_initial_budget_includes_postfit(self):
|
||||
p=project(self.c,{i:60 for i in range(5)},{},{})
|
||||
self.assertEqual(p['status'],'PASS');self.assertEqual(p['projected_gpu_seconds'],178200)
|
||||
def test_remaining_validation_charged(self):
|
||||
p=project(self.c,{i:100 for i in range(5)},{0:[dict(epoch=1,train_epoch_seconds=50)]},{})
|
||||
self.assertEqual(p['folds']['0']['projected'],100+1.5*(199*50+20*3)+120)
|
||||
def test_late_validation_max_used(self):
|
||||
h=[dict(epoch=e,train_epoch_seconds=50,validation_seconds=7 if e==10 else 0) for e in range(1,11)]
|
||||
p=project(self.c,{i:1000 for i in range(5)},{0:h},{})
|
||||
self.assertEqual(p['folds']['0']['projected'],1000+1.5*(190*50+19*7)+120)
|
||||
def test_actual_runaway_stops(self):
|
||||
self.assertEqual(project(self.c,{i:60000 if i==0 else 60 for i in range(5)},{},{})['status'],'STOP_BUDGET')
|
||||
def test_no_single_fold_projection_gate(self):
|
||||
p=project(self.c,{i:1000 for i in range(5)},{0:[dict(epoch=1,train_epoch_seconds=220)]},{1:{},2:{},3:{}})
|
||||
self.assertGreater(p['folds']['0']['projected'],60000);self.assertEqual(p['status'],'PASS')
|
||||
def test_incomplete_history_rejected(self):
|
||||
with self.assertRaises(ValueError):project(self.c,{i:60 for i in range(5)},{0:[dict(epoch=2,train_epoch_seconds=50)]},{})
|
||||
|
||||
|
||||
class SharedWrapperTests(unittest.TestCase):
|
||||
def test_wrapper_changes_only_key_cap_gpu_set(self):
|
||||
import phase3_item_b_capacity as new
|
||||
import phase3_shared_capacity as old
|
||||
for name in ('register_entry','acquire'):
|
||||
s=inspect.getsource(getattr(new,name)).replace("'b'","'c'").replace('item b','item c').replace('cap=178200','cap=205200').replace('GPUS[:3]','GPUS[1:3]')
|
||||
self.assertEqual(ast.dump(ast.parse(s)),ast.dump(ast.parse(inspect.getsource(getattr(old,name)))))
|
||||
for name in ('admission','registry','transfer','release','validate_lease'):
|
||||
self.assertIs(getattr(new,name),getattr(old,name))
|
||||
@contextmanager
|
||||
def mixed(self,reserved=False):
|
||||
import phase3_item_b_capacity as m
|
||||
import phase3_shared_capacity as shared
|
||||
identity=lambda p:dict(pid=p,created=1.)
|
||||
root=Path('/synthetic-1b');oldroot='/synthetic-c'
|
||||
leases={}
|
||||
for n,pid in enumerate([11,12]+([10] if reserved else [])):
|
||||
leases[str(n)]=dict(token=str(n),state='RESERVED' if pid==10 else 'GUARD_BOUND',root=oldroot,fold=n,
|
||||
contract_sha256='old',gpu_uuid=GPUS[1],worker_cpus=[16+n],owner=identity(pid),queue_owner=identity(10),vram_reserve=4*GIB,rss_reserve=4*GIB)
|
||||
original_c=dict(root=oldroot,contract_sha256='old',owner=identity(10),cap=205200)
|
||||
s=dict(devices={g:dict(total=16*GIB,used=4*GIB) for g in GPUS},processes=[dict(gpu_uuid=GPUS[1],pid=p,vram=2*GIB) for p in (101,102)]+[dict(gpu_uuid=GPUS[2],pid=p,vram=GIB) for p in (201,202,203)],ram_total=128*GIB,ram_available=96*GIB,cpu_percent=20,old_pending=True)
|
||||
trees={11:{11,101},12:{12,102},10:{10,11,12,101,102}}
|
||||
with tempfile.TemporaryDirectory() as tmp,ExitStack() as stack:
|
||||
path=Path(tmp)/'registry.json';path.write_text(json.dumps(dict(version=2,entries={'c':original_c},leases=leases)))
|
||||
stack.enter_context(patch.object(shared,'SHARED',Path(tmp)))
|
||||
for name,value in dict(process_identity=lambda p:identity(20),alive=lambda p:True,descendants=lambda p:trees[p['pid']],host_snapshot=lambda:copy.deepcopy(s)).items():stack.enter_context(patch.object(m,name,value))
|
||||
stack.enter_context(patch.object(m.psutil,'Process',lambda p:SimpleNamespace(memory_info=lambda:SimpleNamespace(rss=GIB//4))))
|
||||
stack.enter_context(patch.object(m.os,'sched_getaffinity',lambda p:set(range(16,26))))
|
||||
m.register_entry(root,'new')
|
||||
yield m,root,path,original_c
|
||||
def test_shared_c_and_b_registration_independent(self):
|
||||
with self.mixed() as (m,root,path,old):
|
||||
d=json.loads(path.read_text());self.assertEqual(d['entries']['c'],old);self.assertEqual(d['entries']['b']['cap'],178200)
|
||||
with self.assertRaises(FileExistsError):m.register_entry(root,'new')
|
||||
def test_mixed_pending_c_reservation_blocks_b(self):
|
||||
with self.mixed(True) as (m,root,path,old):
|
||||
before=json.loads(path.read_text());l,_=m.acquire(root,1,'new');self.assertIsNone(l)
|
||||
self.assertEqual(json.loads(path.read_text()),before)
|
||||
def test_mixed_two_visible_allows_only_one_new(self):
|
||||
with self.mixed() as (m,root,path,old):
|
||||
l,_=m.acquire(root,1,'new');self.assertIsNotNone(l);self.assertEqual(l['state'],'RESERVED');self.assertEqual(l['gpu_uuid'],GPUS[1])
|
||||
second,_=m.acquire(root,0,'new');self.assertIsNone(second)
|
||||
self.assertEqual(json.loads(path.read_text())['entries']['c'],old)
|
||||
|
||||
|
||||
class QueueTests(unittest.TestCase):
|
||||
def exercise(self,failed=False):
|
||||
import phase3_item_b_queue as q
|
||||
from phase2_h1_worker_v2 import save
|
||||
with tempfile.TemporaryDirectory() as tmp,ExitStack() as stack:
|
||||
root=Path(tmp);save(root/'contract.json',dict(synthetic=True));launched=[]
|
||||
c=dict(gpu_cap=178200,postfit_gpu_reserve=7200,preparation_charge_gpu_seconds=600,runaway_gpu_seconds=60000,jobs=[dict(validation_seconds_bound=1) for _ in range(5)])
|
||||
class Process:
|
||||
def __init__(self,cmd,**kw):
|
||||
self.pid=999991;self.returncode=int(failed);i=int(cmd[3]);launched.append(i);j=root/'jobs'/str(i);(j/'worker').mkdir()
|
||||
save(j/'complete.json',dict(status='STOPPED_OR_FAILED' if failed else 'PASS'));p=j/'worker/last-epoch200.pt';p.write_bytes(b'synthetic')
|
||||
save(j/'worker/result.json',dict(fold=i,checkpoint=str(p),checkpoint_sha256=q.sha(p),completed_epochs=200));save(j/'worker/standardization.json',{})
|
||||
def poll(self):return self.returncode
|
||||
def wait(self):return self.returncode
|
||||
def terminate(self):pass
|
||||
def kill(self):pass
|
||||
replacements=dict(config=lambda r:(c,{}),validate_review=lambda r:None,register_entry=lambda *a:None,
|
||||
acquire=lambda r,i,d:(dict(token=str(i),gpu_uuid=GPUS[1],worker_cpus=[20]),{}),transfer=lambda *a:None,
|
||||
release=lambda *a:None,process_identity=lambda p:dict(pid=p,created=0),freeze=lambda *a:None)
|
||||
for k,v in replacements.items():stack.enter_context(patch.object(q,k,v))
|
||||
stack.enter_context(patch.object(q.subprocess,'Popen',Process));stack.enter_context(patch.object(q.time,'sleep',lambda t:None))
|
||||
stack.enter_context(patch.object(q.signal,'signal',lambda *a:None));stack.enter_context(patch.object(q.psutil,'disk_usage',return_value=SimpleNamespace(free=200*GIB)))
|
||||
stack.enter_context(patch.dict(q.os.environ,PYTORCH_CUDA_ALLOC_CONF='expandable_segments:True'))
|
||||
result=q.run(root);d=q.read(root/'complete.json')
|
||||
if failed:self.assertEqual(result,1);self.assertEqual(launched,[0]);self.assertFalse((root/'models.json').exists())
|
||||
else:
|
||||
self.assertEqual(result,0);self.assertEqual(launched,[0,1,2,3,4]);self.assertEqual(len(q.read(root/'models.json')),5)
|
||||
self.assertEqual(d['status'],'TRAIN_COMPLETE_POSTFIT_PENDING')
|
||||
self.assertFalse(d['test_targets_read']);self.assertFalse(d['prediction_started']);self.assertFalse(d['scoring_started'])
|
||||
def test_five_fits_stop_before_postfit(self):self.exercise()
|
||||
def test_failure_no_retry_or_test(self):self.exercise(True)
|
||||
|
||||
|
||||
Reference in new issue
Block a user