File size: 33,294 Bytes
6158a59 ebdefe7 9e36559 6158a59 24e5e23 6158a59 0d4b28f 24e5e23 ebdefe7 6158a59 9e36559 6158a59 24e5e23 ebdefe7 9e36559 6158a59 9e36559 6158a59 9e36559 6158a59 0d4b28f ebdefe7 24e5e23 9e36559 6158a59 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 | """SYNAPSE command line entry point."""
import argparse
import json
from pathlib import Path
from .config import SynapseConfig, ARTIFACTS
def main(argv=None):
parser=argparse.ArgumentParser(description='SYNAPSE integrated connectome research lab')
parser.add_argument('--output',type=Path,default=ARTIFACTS)
parser.add_argument('--seed',type=int,default=7)
parser.add_argument('--frontier-all',action='store_true',help='activate all frontier cognitive, planning, self-modeling, and neuromorphic features')
parser.add_argument('--compute-threads',type=int,help='CPU thread budget, bounded by available logical CPUs')
parser.add_argument('--neural-graft',action='store_true',help='enable the trainable bounded neural residual')
parser.add_argument('--active-training',action='store_true',help='fit the residual from actual live exploratory outcomes')
parser.add_argument('--programmed-connectome',action='store_true',help='bind the verified 128-neuron MaleCNS sparse prior')
parser.add_argument('--connectome-mode',choices=['proxy','prefer_authentic','strict_authentic'],help='connectome source mode')
size=parser.add_mutually_exclusive_group()
size.add_argument('--nodes',type=int,help='legacy neurons per source')
size.add_argument('--total-neurons',type=int,help='exact total live actor neurons (default: 6000)')
parser.add_argument('--task',default='forage')
sub=parser.add_subparsers(dest='command',required=True)
sub.add_parser('status')
for name in ('run','train'):
p=sub.add_parser(name); p.add_argument('--steps',type=int,default=100); p.add_argument('--save')
p=sub.add_parser('serve'); p.add_argument('--port',type=int,default=8765); p.add_argument('--host',default='127.0.0.1')
p=sub.add_parser('study'); p.add_argument('--config',type=Path); p.add_argument('--name',default='cli-study')
p=sub.add_parser('verify'); p.add_argument('path',type=Path)
p=sub.add_parser('capabilities'); p.add_argument('--id'); p.add_argument('--payload',default='{}')
p=sub.add_parser('patch'); p.add_argument('operation',choices=['make','apply']); p.add_argument('file',type=Path); p.add_argument('--checkpoint'); p.add_argument('--save',default='patched')
p=sub.add_parser('autoprogram'); p.add_argument('--diagnostic-episodes',type=int,default=3); p.add_argument('--eval-episodes',type=int,default=3); p.add_argument('--min-gain',type=float,default=-0.05); p.add_argument('--save')
p=sub.add_parser('transplant'); p.add_argument('--donor-task',default='forage'); p.add_argument('--test-task',default='delayed_match'); p.add_argument('--episodes',type=int,default=3)
sub.add_parser('metamorphic')
sub.add_parser('spectral')
p=sub.add_parser('hyperprogram'); p.add_argument('--target',default='motor.weight'); p.add_argument('--edits',type=int,default=16)
sub.add_parser('astrocyte')
sub.add_parser('free_energy')
sub.add_parser('manifold')
p=sub.add_parser('flagship'); p.add_argument('--training-steps',type=int,default=15); p.add_argument('--generations',type=int,default=3)
sub.add_parser('spiking')
sub.add_parser('atlas')
sub.add_parser('efficiency')
p=sub.add_parser('connectome'); p.add_argument('--query',default='motor')
sub.add_parser('neurogenesis')
sub.add_parser('homeostasis')
p=sub.add_parser('mimomix'); p.add_argument('--ponder',type=int,default=4)
p=sub.add_parser('ecology'); p.add_argument('--task',default='ecosystem')
sub.add_parser('continual-backprop')
p=sub.add_parser('sleep-train'); p.add_argument('--sleep-steps',type=int,default=10)
sub.add_parser('fast-memory')
p=sub.add_parser('multistep-plan'); p.add_argument('--horizon',type=int,default=5)
sub.add_parser('curriculum-verify')
sub.add_parser('causal-growth')
sub.add_parser('connectome-datasets')
p=sub.add_parser('multi-connectome'); p.add_argument('--nodes-per-source', type=int, default=32)
p=sub.add_parser('scale-neurons'); p.add_argument('--targets', type=int, nargs='+', default=[64, 128, 256, 512])
p=sub.add_parser('rsi-verify'); p.add_argument('--patch-type', default='config')
p=sub.add_parser('world-model-cem'); p.add_argument('--cem-horizon', type=int, default=5)
p=sub.add_parser('connectome-assay'); p.add_argument('--assay-trials', type=int, default=2)
p=sub.add_parser('coconut-think'); p.add_argument('--think-depth', type=int, default=4)
p=sub.add_parser('diffusion-repair'); p.add_argument('--repair-iters', type=int, default=3)
p=sub.add_parser('dpo-align'); p.add_argument('--pairs', type=int, default=5)
p=sub.add_parser('unified-eval'); p.add_argument('--eval-task', default='forage')
p=sub.add_parser('flagship-milestone'); p.add_argument('--milestone-steps',type=int,default=20)
p=sub.add_parser('controlled-study'); p.add_argument('--steps-per-stage',type=int,default=15); p.add_argument('--eval-episodes',type=int,default=1)
sub.add_parser('uncertainty-calibrate')
sub.add_parser('self-model-audit')
sub.add_parser('reusable-skills')
p=sub.add_parser('neural-comm'); p.add_argument('--steps', type=int, default=15)
p=sub.add_parser('world4d'); p.add_argument('--steps', type=int, default=20)
p=sub.add_parser('tool-arena'); p.add_argument('--steps', type=int, default=10)
sub.add_parser('benchmarks')
sub.add_parser('motifs')
sub.add_parser('meta-plasticity')
sub.add_parser('ablation')
sub.add_parser('checkpoint-migration')
p=sub.add_parser('omnibus'); p.add_argument('--steps-per-task',type=int,default=30); p.add_argument('--tasks',nargs='+',default=None); p.add_argument('--save-report',type=Path)
p=sub.add_parser('evidence'); p.add_argument('--save-report',type=Path)
p.add_argument('--receipt',type=Path,help='read and verify an existing measured adaptation receipt')
p.add_argument('--seeds',type=int,nargs='+',default=[7,11,19])
p.add_argument('--tasks',nargs='+',default=['forage','push'])
p.add_argument('--support-steps',type=int,default=64)
p.add_argument('--query-steps',type=int,default=64)
p.add_argument('--study-nodes',type=int,default=16)
p.add_argument('--name',help='exclusive output folder name; defaults to a timestamp')
p=sub.add_parser('adaptation-verify'); p.add_argument('path',type=Path)
p=sub.add_parser('pretrain'); p.add_argument('--steps',type=int,default=256); p.add_argument('--updates',type=int,default=64); p.add_argument('--batch-size',type=int,default=128); p.add_argument('--workers',type=int,default=2); p.add_argument('--save',default='pretrained-graft')
p=sub.add_parser('live-rsi'); p.add_argument('--steps',type=int,default=24); p.add_argument('--candidates',type=int,default=2); p.add_argument('--save')
sub.add_parser('connectome-inventory')
p=sub.add_parser('planning-study'); p.add_argument('--name',default='planning-pilot')
p.add_argument('--seeds',type=int,nargs='+',default=[23,29,31])
p.add_argument('--tasks',nargs='+',default=['forage','push'])
p.add_argument('--support-steps',type=int,default=64)
p.add_argument('--query-steps',type=int,default=64)
p.add_argument('--horizons',type=int,nargs='+',default=[1,3])
p=sub.add_parser('planning-verify'); p.add_argument('path',type=Path)
args=parser.parse_args(argv)
try:
total=args.total_neurons if args.total_neurons is not None else (6000 if args.nodes is None else None)
if getattr(args, 'frontier_all', False):
config=SynapseConfig.frontier_apex(total_neurons=total,seed=args.seed,task=args.task)
else:
config=SynapseConfig(seed=args.seed,nodes_per_source=100 if args.nodes is None else args.nodes,total_neurons=total,task=args.task)
if args.command=='serve' and args.nodes is None and not getattr(args, 'frontier_all', False):
from dataclasses import replace
import os
config=replace(config,
neural_graft=True,
active_training=True,
programmed_connectome=True,
compute_threads=args.compute_threads if args.compute_threads is not None else min(4,os.cpu_count() or 1),
connectome_mode=args.connectome_mode or 'prefer_authentic')
elif (args.compute_threads is not None or args.neural_graft or args.active_training
or args.programmed_connectome or args.connectome_mode is not None or args.command=='pretrain'):
from dataclasses import replace
import os
enable_graft = (config.neural_graft or args.neural_graft or args.active_training
or args.programmed_connectome or args.command=='pretrain')
config=replace(config,
neural_graft=enable_graft,
active_training=config.active_training or args.active_training,
programmed_connectome=config.programmed_connectome or args.programmed_connectome,
connectome_mode=args.connectome_mode or config.connectome_mode,
compute_threads=args.compute_threads if args.compute_threads is not None else min(4,os.cpu_count() or 1))
demo_nodes=config.total_neurons or config.nodes_per_source
if args.command=='serve':
from .server import serve
serve(args.host,args.port,config=config,output_root=args.output); return 0
if args.command=='connectome-inventory':
from .connectome_assets import get_source_inventory
result={'sources':get_source_inventory()}
elif args.command=='capabilities':
from .capabilities import catalog,invoke
result=invoke(args.id,json.loads(args.payload)) if args.id else catalog()
elif args.command in ('study','verify'):
from .experiments import run_study,verify_study
from .server import safe_name
result=verify_study(args.path) if args.command=='verify' else run_study(args.output/'studies'/safe_name(args.name),json.loads(args.config.read_text()) if args.config else {})
elif args.command=='adaptation-verify':
from .adaptation_study import verify_adaptation_study
result=verify_adaptation_study(args.path)
elif args.command=='planning-verify':
from .planning_study import verify_planning_study
result=verify_planning_study(args.path)
elif args.command=='planning-study':
from .planning_study import run_planning_study
from .server import safe_name
result=run_planning_study(args.output/'planning-studies'/safe_name(args.name),
{'seeds':args.seeds,'tasks':args.tasks,'support_steps':args.support_steps,
'query_steps':args.query_steps,'nodes_per_source':config.nodes_per_source,'horizons':args.horizons})
elif args.command=='evidence':
from .adaptation_study import run_adaptation_study,verify_adaptation_study,format_adaptation_report
from .server import safe_name
if args.receipt:
verification=verify_adaptation_study(args.receipt)
if not verification['ok']:
print(json.dumps(verification,indent=2)); return 1
path=args.receipt/'receipt.json' if args.receipt.is_dir() else args.receipt
result=json.loads(path.read_text(encoding='utf-8'))
else:
from time import time_ns
name=safe_name(args.name or f'pilot-{time_ns()}')
result=run_adaptation_study(args.output/'adaptation-studies'/name,{'seeds':args.seeds,'tasks':args.tasks,'support_steps':args.support_steps,'query_steps':args.query_steps,'nodes_per_source':args.study_nodes})
if args.save_report:
args.save_report.write_text(format_adaptation_report(result),encoding='utf-8')
else:
from .runtime import LabRuntime
from .server import integer,safe_name
runtime=LabRuntime(config,output_root=args.output)
if args.command=='status': result=runtime.snapshot()
elif args.command=='pretrain':
from .active_training import pretrain_runtime
result=pretrain_runtime(runtime,steps=args.steps,updates=args.updates,batch_size=args.batch_size,workers=args.workers)
if args.save: result['checkpoint']=runtime.save(safe_name(args.save))
elif args.command=='live-rsi':
from .live_programming import run_live_rsi
result=run_live_rsi(runtime.agent,task=config.task,steps=args.steps,candidates=args.candidates,seed=args.seed)
if args.save: result['checkpoint']=runtime.save(safe_name(args.save))
elif args.command in ('run','train'):
result=runtime.step(integer(args.steps,'steps',1,10000),learn=args.command=='train')
if args.save: result={'state':result,'checkpoint':runtime.save(safe_name(args.save))}
elif args.command=='patch':
from .patches import make_patch,apply_patch
if args.checkpoint: runtime.load(safe_name(args.checkpoint))
payload=json.loads(args.file.read_text())
if args.operation=='make': result=make_patch(runtime.agent,payload.get('edits',payload) if isinstance(payload,dict) else payload)
else: result={'receipt':apply_patch(runtime.agent,payload),'checkpoint':runtime.save(safe_name(args.save))}
elif args.command=='autoprogram':
from .auto_programmer import execute_closed_loop_repair
result=execute_closed_loop_repair(runtime.agent,task=args.task,num_diagnostic_episodes=args.diagnostic_episodes,num_eval_episodes=args.eval_episodes,min_gain=args.min_gain,seed=args.seed)
if args.save: result={'report':result,'checkpoint':runtime.save(safe_name(args.save))}
elif args.command=='transplant':
from .transplantation import extract_donor_subcircuit,run_transplantation_experiment
donor=extract_donor_subcircuit(runtime.agent,source_task=args.donor_task)
result=run_transplantation_experiment(runtime.agent,donor=donor,test_task=args.test_task,episodes=args.episodes,seed=args.seed)
elif args.command=='metamorphic':
from .metamorphic import run_metamorphic_verification_suite
result=run_metamorphic_verification_suite(runtime.agent)
elif args.command=='spectral':
from .spectral import compute_connectome_harmonics,project_spectral_activity
harmonics=compute_connectome_harmonics(runtime.agent.graph,num_modes=8)
proj=project_spectral_activity(runtime.agent.brain.r,harmonics)
result={'eigenvalues':harmonics['eigenvalues'],'projection':proj}
elif args.command=='hyperprogram':
from .hypernetwork import WeightSpaceHypernetwork,compile_hypernetwork_patch
hypernet=WeightSpaceHypernetwork()
diag={'mean_return':0.2,'total_steps':30,'mean_graph_activity':float(runtime.agent.brain.r.mean()),'mean_approach_valence':0.1,'mean_avoid_valence':-0.1,'action_mean':runtime.last_action.tolist(),'task':runtime.config.task}
result=compile_hypernetwork_patch(runtime.agent,hypernet,diag,target_surface=args.target,max_edits=args.edits)
elif args.command=='astrocyte':
result=runtime.agent.astrocytes.summary()
elif args.command=='free_energy':
result=runtime.agent.predictive.state.to_dict()
elif args.command=='manifold':
from .manifold_alignment import align_subcircuit_gromov_wasserstein
import numpy as np
adj=np.array([[0,1,0,1],[1,0,1,0],[0,1,0,1],[1,0,1,0]],dtype=np.float64)
result=align_subcircuit_gromov_wasserstein(adj,adj)
elif args.command=='flagship':
from .flagship import FlagshipExperimentRunner
runner=FlagshipExperimentRunner(seed=args.seed)
result=runner.run_full_experiment(training_steps=args.training_steps,evolution_generations=args.generations)
elif args.command=='spiking':
from .spiking import benchmark_spiking_vs_rate
result=benchmark_spiking_vs_rate(num_neurons=demo_nodes,steps=50,seed=args.seed)
elif args.command=='atlas':
from .circuit_atlas import CausalCircuitAtlas
atlas=CausalCircuitAtlas(runtime.agent)
result=atlas.build_full_atlas()
elif args.command=='efficiency':
from .efficiency import NeuromorphicEfficiencyBenchmark
bench=NeuromorphicEfficiencyBenchmark(num_neurons=demo_nodes,seed=args.seed)
result=bench.run_benchmark_suite(steps=50)
elif args.command=='connectome':
from .connectome_compiler import BiologicalConnectomeCompiler
comp=BiologicalConnectomeCompiler()
m=comp.compile_subcircuit(query=args.query,max_neurons=demo_nodes)
null=comp.generate_degree_preserving_null(m)
result=comp.benchmark_biological_vs_null(m,null)
elif args.command=='neurogenesis':
result=runtime.agent.neurogenesis.summary()
elif args.command=='homeostasis':
result=runtime.agent.homeostasis.summary()
elif args.command=='mimomix':
import torch
latent = torch.randn(64)
mod, trace = runtime.agent.mimomix.deliberate(latent, free_energy=0.8, surprisal=0.6, threat=0.2)
import dataclasses
result={'modification': mod.tolist(), 'trace': dataclasses.asdict(trace)}
elif args.command=='ecology':
from .complex_world import ComplexWorld
world = ComplexWorld(seed=args.seed)
obs = world.reset(task=args.task)
import numpy as np
next_obs, reward, done, info = world.step(np.array([0.5, 0.0, 0.0, 0.0, 0.0], dtype=np.float32))
result={'task': world.task, 'predators': len(world.predators), 'friction_zones': len(world.terrain_zones), 'gates': len(world.dynamic_gates), 'ambient_light': info.get('ambient_light', 1.0), 'reward': float(reward)}
elif args.command=='continual-backprop':
from .continual_backprop import benchmark_continual_backprop
result=benchmark_continual_backprop(runtime.agent, steps=30, seed=args.seed)
elif args.command=='sleep-train':
from .sleep import SleepConsolidationEngine
sleep=SleepConsolidationEngine()
import numpy as np
for _ in range(12):
sleep.record_experience(np.random.uniform(0,1,155).astype(np.float32), np.zeros(5,dtype=np.float32), reward=1.0)
result=sleep.run_sleep_cycle(runtime.agent, sleep_steps=args.sleep_steps)
elif args.command=='fast-memory':
from .fast_weight_memory import UniversalFastWeightMemory
import torch
fmem=UniversalFastWeightMemory(in_dim=32, out_dim=32, seed=args.seed)
res=fmem.step_and_update(torch.randn(32), surprisal=0.6)
result={'step': res['step'], 'inner_loss': res['inner_loss'], 'summary': fmem.summary()}
elif args.command=='multistep-plan':
from .multistep_planner import MultistepRecurrentPlanner, compare_planning_horizons
planner=MultistepRecurrentPlanner(seed=args.seed)
import numpy as np
obs=np.random.uniform(0,1,155).astype(np.float32)
result=compare_planning_horizons(planner, obs, np.zeros(5,dtype=np.float32), horizons=[1, 3, args.horizon])
elif args.command=='curriculum-verify':
from .verified_curriculum import VerifiableCurriculumEngine
curric=VerifiableCurriculumEngine(base_seed=args.seed)
ch=curric.generate_procedural_challenge()
sub=curric.check_subgoals((0.0, 0.0))
rec=curric.complete_episode(success=True)
result={'stage': curric.current_stage, 'proof_hash': rec.proof_hash, 'summary': curric.summary()}
elif args.command=='causal-growth':
from .causal_growth import CausalConnectomeEvolutionEngine
c_eng=CausalConnectomeEvolutionEngine(seed=args.seed)
mut=c_eng.generate_candidate_mutation(runtime.agent)
rec=c_eng.evaluate_interventional_mutation(runtime.agent, mut, test_episodes=2, horizon=15)
result={'mutation': mut.description, 'causal_delta': rec.causal_delta, 'net_fitness': rec.net_fitness, 'accepted': rec.accepted}
elif args.command=='connectome-datasets':
from .connectome_datasets import list_connectome_datasets
result = {'status': 'ok', 'total_datasets': 8, 'datasets': list_connectome_datasets()}
elif args.command=='multi-connectome':
from .multi_connectome import MultiConnectomeHub
import torch
hub = MultiConnectomeHub(nodes_per_source=args.nodes_per_source, seed=args.seed)
inp = torch.randn(len(hub.brain.sensory_idx))
r = hub(inp)
result = hub.get_summary()
elif args.command=='scale-neurons':
from .multi_connectome import benchmark_neuron_scaling
result = benchmark_neuron_scaling(target_counts=args.targets, steps_per_benchmark=5, seed=args.seed)
elif args.command=='rsi-verify':
from .rsi_verifier import BoundedRSIEngine
rsi_eng = BoundedRSIEngine(seed=args.seed)
cand = rsi_eng.generate_candidate(patch_type=args.patch_type)
rec = rsi_eng.verify_and_eval(cand, num_support_trials=3, num_query_trials=3)
result = {'candidate': cand.patch_id, 'passed_all': rec.passed_all, 'final_stage': rec.final_stage.value, 'lcb_score': rec.lcb_score, 'receipt_hash': rec.receipt_hash, 'summary': rsi_eng.get_summary()}
elif args.command=='world-model-cem':
from .world_model_v2 import RecurrentLatentWorldModelEnsemble, CrossEntropyMethodPlanner
wm = RecurrentLatentWorldModelEnsemble(num_members=3, state_dim=32, action_dim=5, obs_dim=155, seed=args.seed)
planner = CrossEntropyMethodPlanner(wm, action_dim=5, default_horizon=args.cem_horizon, num_samples=8, num_elites=2, cem_iterations=2, seed=args.seed)
import numpy as np
obs = np.random.uniform(0, 1, 155).astype(np.float32)
plan_res = planner.plan(obs, np.zeros(5, dtype=np.float32))
result = {'plan_mode': plan_res['plan_mode'], 'shield_triggered': plan_res['shield_triggered'], 'best_score': plan_res['best_score'], 'exploitation': None, 'realized_outcomes': 0}
elif args.command=='connectome-assay':
from .connectome_compiler import BiologicalConnectomeCompiler
from .connectome_controls import ControlledConnectomeAssay
comp = BiologicalConnectomeCompiler()
bio = comp.compile_subcircuit(max_neurons=16, min_weight=1.0)
assay = ControlledConnectomeAssay(compiler=comp, seed=args.seed)
result = assay.run_causal_assay(bio, steps=10, trials=args.assay_trials)
elif args.command=='coconut-think':
from .latent_reasoner import ContinuousLatentReasoner
import torch
reasoner = ContinuousLatentReasoner(latent_dim=32, action_dim=5, max_depth=args.think_depth)
z = torch.randn(32)
act, trace = reasoner.forward_continuous_thinking(z)
result = {'recurrent_depth': trace.recurrent_depth, 'ponder_cost': trace.ponder_cost, 'thought_norm': trace.latent_thought_norm, 'action': act.squeeze(0).tolist()}
elif args.command=='diffusion-repair':
from .diffusion_repair import DiffusionRepairSidecar
import numpy as np
sidecar = DiffusionRepairSidecar(feature_dim=8, max_repair_iterations=args.repair_iters, diffusion_timesteps=5)
cand = np.array([1.5, -2.0, 0.5, 0.2, -0.8, 1.2, 0.1, -0.4], dtype=np.float32)
def verifier(vec):
failing = [i for i, v in enumerate(vec) if abs(v) > 1.0]
return len(failing) == 0, failing, {}
rec = sidecar.repair_candidate_vector(cand, [0, 1, 5], verifier)
result = {'initial_valid': rec.initial_valid, 'final_valid': rec.final_valid, 'iterations_used': rec.iterations_used, 'repaired_diff': rec.repaired_diff}
elif args.command=='dpo-align':
from .preference_dpo import MachineVerifiableDPOEngine
import numpy as np
dpo_eng = MachineVerifiableDPOEngine(obs_dim=32, action_dim=5, beta=0.1)
for _ in range(args.pairs):
obs = np.random.randn(32).astype(np.float32)
w = np.random.uniform(-0.5, 0.5, size=5).astype(np.float32)
l = np.random.uniform(-1.0, 1.0, size=5).astype(np.float32)
dpo_eng.record_preference(obs, w, l, criterion="safety", delta_utility=1.5)
train_out = dpo_eng.train_dpo_step(batch_size=4)
result = {'loss_dpo': train_out.get('loss_dpo', 0.0), 'implicit_reward_margin': train_out.get('implicit_reward_margin', 0.0), 'buffer_size': train_out.get('buffer_size', 0)}
elif args.command=='unified-eval':
from .unified_eval import UnifiedEvaluator
import numpy as np
evaluator = UnifiedEvaluator()
preds = np.array([0.9, 0.8, 0.2, 0.1, 0.85])
targets = np.array([1, 1, 0, 0, 1])
card = evaluator.evaluate_task_run(args.eval_task, returns=[15.0, 18.0], execution_times=[0.05, 0.06], predictions=preds, ground_truth=targets, failures=0)
result = card.to_dict()
elif args.command=='flagship-milestone':
from .flagship_benchmark import FlagshipContinualBenchmark
bench=FlagshipContinualBenchmark(steps_per_stage=args.milestone_steps, seed=args.seed)
result=bench.run_benchmark()
elif args.command=='controlled-study':
from .controlled_study import ControlledSixArmStudy
study = ControlledSixArmStudy(steps_per_stage=args.steps_per_stage, eval_episodes_per_task=args.eval_episodes, seed=args.seed)
result = study.run_study()
elif args.command=='uncertainty-calibrate':
from .uncertainty_calibration import TripartiteUncertaintyDecomposer
import numpy as np
dec = TripartiteUncertaintyDecomposer(state_dim=16, action_dim=5, seed=args.seed)
z_val = np.random.randn(20, 16).astype(np.float32)
a_val = np.random.uniform(-1, 1, (20, 5)).astype(np.float32)
r_val = np.random.randn(20).astype(np.float32)
result = dec.calibrate_on_held_out(z_val, a_val, r_val)
elif args.command=='self-model-audit':
from .self_model import EfferenceCopyComparator, ControlMode
import numpy as np
comp = EfferenceCopyComparator(sensory_dim=155, action_dim=5, seed=args.seed)
s0 = np.random.randn(155).astype(np.float32)
a0 = np.random.uniform(-0.5, 0.5, 5).astype(np.float32)
s1 = s0 + 0.1 * np.pad(a0, (0, 150))
rec = comp.process_transition(s0, a0, s1, learn=True)
result = {'receipt': rec.to_dict(), 'summary': comp.summary()}
elif args.command=='reusable-skills':
from .reusable_skills import ReusableSkillLibrary
lib = ReusableSkillLibrary(seed=args.seed)
result = lib.summary()
elif args.command=='omnibus':
from .omnibus_mode import OmnibusRunner
runner = OmnibusRunner(
total_neurons=total or 2000,
seed=args.seed,
steps_per_task=args.steps_per_task,
curriculum_tasks=args.tasks,
output_dir=args.output
)
receipt = runner.run_full_curriculum()
result = receipt.to_dict()
if args.save_report:
args.save_report.write_text(json.dumps(result, indent=2), encoding='utf-8')
elif args.command=='neural-comm':
from .neural_comm_bench import SancfMatrixBenchmark
bench = SancfMatrixBenchmark(num_steps=args.steps, seed=args.seed)
m_res = bench.run_benchmark()
nw_res = bench.run_neurowave_assay()
import dataclasses
result = {
'sancf_matrix': [dataclasses.asdict(a) for a in m_res.arm_results],
'neurowave_modes': [dataclasses.asdict(m) for m in nw_res.modes],
'neurowave_controls': [dataclasses.asdict(c) for c in nw_res.controls],
}
elif args.command=='world4d':
from .world4d import World4DEnvironment
env = World4DEnvironment(seed=args.seed)
env.reset()
import numpy as np
step_res = env.step(np.array([0.5, 0.0, 0.3, 0.0, 0.0], dtype=np.float32))
result = {'obs_dim': len(step_res[0]), 'reward': step_res[1], 'done': step_res[2], 'snapshot': env.snapshot()}
elif args.command=='tool-arena':
from .digital_tool_arena import DigitalToolArena, ToolActionType
arena = DigitalToolArena(seed=args.seed)
t_info = arena.reset()
step_res = arena.step(ToolActionType.LIST, {})
import dataclasses
result = {'task': dataclasses.asdict(t_info), 'step_result': dataclasses.asdict(step_res), 'snapshot': arena.snapshot()}
elif args.command=='benchmarks':
from .benchmark_suite import BenchmarkRunner, ResearchScorecard
runner = BenchmarkRunner()
b_results = runner.run_all(seed=args.seed)
scorecard = ResearchScorecard().generate_scorecard(b_results)
result = {'overall_pass_rate': scorecard.overall_pass_rate, 'scorecard': scorecard.summary_markdown}
elif args.command=='motifs':
from .causal_connectome_assay import CausalConnectomeAssay
assay = CausalConnectomeAssay(num_nodes=200, seed=args.seed)
c_res = assay.run_full_assay()
result = {'motifs_count': len(c_res.motifs), 'interventions_count': len(c_res.interventions), 'notes': c_res.disclaimer}
elif args.command=='meta-plasticity':
from .plasticity_rule_learner import PlasticityRuleLearner
learner = PlasticityRuleLearner(seed=args.seed)
l_res = learner.train_and_evaluate(meta_epochs=3)
result = {'learned_rule_score': l_res.learned_rule_score, 'hebbian_baseline_score': l_res.hebbian_baseline_score}
elif args.command=='ablation':
from .ablation_engine import FactorialAblationEngine
ablation = FactorialAblationEngine(num_seeds=2)
a_res = ablation.run_study()
import dataclasses
result = {'summary': a_res.summary_findings, 'conditions': [dataclasses.asdict(c) for c in a_res.conditions]}
elif args.command=='checkpoint-migration':
from .checkpoint_migration import CheckpointMigrationSimulator
import dataclasses
mig = CheckpointMigrationSimulator()
dummy_state = {'config': {'seed': args.seed}, 'activations': [0.1, 0.2, 0.3], 'step_count': 100}
res = mig.simulate_migration(dummy_state)
result = dataclasses.asdict(res)
if args.command in ('world-model-cem','dpo-align','unified-eval','coconut-think','multistep-plan','rsi-verify'):
result={**result,'measured':False,'evidence_kind':'illustrative',
'scope':'standalone synthetic inputs; no measured embodied performance; use planning-study or evidence for actual world trials'}
print(json.dumps(result,indent=2,allow_nan=False,default=str))
return 1 if args.command in ('verify','adaptation-verify','planning-verify') and not result.get('ok',False) else 0
except (ValueError,TypeError,KeyError,OSError,RuntimeError) as exc:
parser.exit(2,f'Error: {exc}\n')
if __name__=='__main__': raise SystemExit(main())
|