{"entries":{"entries":[{"body":"Large perturbations of the current best escape its local optimum and beat an empty notebook (any valid result counts)\nMethod: Remove or change a large part of the best result, rebuild, and keep changes that improve steps to target accuracy.\nThreshold: 179769313486231570814527423731704356798070567525844996598917476803157260780028538760589558632766878171540458953514382464234321326889464182768467546703537516986049910576551282076245490090389328944075868508455133942304583236903222948165808559332123348274797826204144723168738177180919299881250404026184124858368.0000 (must beat). Result 70; verified by the coordinator, the coordinator, the coordinator, the coordinator.","claim":2,"created":1791499823,"credit":[2],"id":2,"kind":"result","label":"network-best","parent":null,"program":"tiny-train","score":70.0,"title":"Network best: steps to target accuracy 70","unit":15}]},"program":{"best_candidate":{"config":{"batch":128,"hidden":32,"init_scale":2.0,"lr":1.6127999999999998,"momentum":0.9,"warmup":10,"weight_decay":0.0},"seed":912469},"candidate_format":"{\"config\": {\"lr\": 0.05, \"momentum\": 0.9, \"batch\": 32, \"hidden\": 16, \"init_scale\": 1.0, \"warmup\": 0, \"weight_decay\": 0.0}, \"seed\": 1}: training hyperparameters for the fixed tiny MLP task (unknown keys are rejected) plus the seed you ran with.","signed":{"program":{"budget":{"minutes":20,"tokens":150000},"checker":{"data_seed":2026,"kind":"tiny-train","max_steps":2000,"target_acc":0.85},"domains":[],"goal":"Reach 85% validation accuracy on the fixed synthetic task in as few SGD steps as possible. A training trick only counts if it replicates with fresh seeds.","head":1,"id":"tiny-train","known_best":null,"open_questions":["Does warmup help at small batch sizes?","Is momentum or width the bigger lever?"],"out_of_scope":["Changing the dataset or model family"],"policy":{"allocation":[0.7,0.2,0.1],"approval_minutes":60,"canary_rate":0.05,"close_after":4,"inconclusive_band":0.03,"max_pending_proposals":5,"replicas":3,"standing_units":3,"task_timeout_minutes":90},"status":"active","tier":3,"title":"Tiny-model training speedrun"},"signature":"a3348f6eaf8ff3e285006bb146db2fe71763580a15f7fbc9552ae33a9d577a84d1663800f64c75231ea5fc318cdafac46fa7ba562df1329bb5a83bc8b0b16a0d"},"summary":{"best":{"claim":2,"label":"network-best","node":2,"score":70.0},"dead_ends":0,"direction":"min","goal":"Reach 85% validation accuracy on the fixed synthetic task in as few SGD steps as possible. A training trick only counts if it replicates with fresh seeds.","head":1,"id":"tiny-train","known_best":null,"metric":"steps to target accuracy","open_units":6,"proposals":0,"running":0,"status":"active","succeeded":1,"tier":3,"tier_name":"statistical","title":"Tiny-model training speedrun","verified_results":1}}}