summaryrefslogtreecommitdiff
path: root/docs/campaign/CASCADE_ABLATION_PLAN.md
AgeCommit message (Collapse)Author
2026-07-16RESULT 31: decoupling — tail win is 101% beta, centered CE-neutral at 42M ↵Yuren Hao
tail; plain-full-epoch + 72M A/B queued post-crown; beta ablation auto-triggered Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-16RESULT 30: quantization Delta-vs-Delta — zero-tax at 8-bit operating point ↵Yuren Hao
for BOTH; EP-specific tax (1.4-8x) below it; beta-buyback prediction designed Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-16RESULT 29: qcomp8 tax-free (+0.0018) = T64 8-bit compute green light; ↵Yuren Hao
--centmirror ships (cos 1.000000000, 1.72x -> 1.39x) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-16RESULT 28 caveat: fp32-EP control insufficient — BP mirror is the decision ↵Yuren Hao
metric (user correction) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-16RESULT 28: Stage-0 gate #1 verdict (naked resident cells need >=10-12b; T64 ↵Yuren Hao
word-streaming exempt; qcomp = the decisive arm) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-15RESULT 27: K-saturation across training stages — K3=K8 at s5k/50k/100k; ↵Yuren Hao
fixed-point bias (not truncation) grows; answers Alexi's more-nudges challenge Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-15ERRATUM RESULT 21: original best 3.7117 was @104.8k not @185k — 90k-step ↵Yuren Hao
wall-1 plateau before the wall-2 blow; figure + deck rebuilt Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-15RESULT 26: C512 ZERO-GAP SEALED — stage1b_cent 1.2334 vs BP 1.2311 ↵Yuren Hao
(+0.0023, inside seed band); the 0.050 gap was recipe/noise Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-15External notes: sliding-mode clamping (Stage-0 injection rule), ↵Yuren Hao
saturated-unit EP-BPTT divergence (smooth-gate methods rule), momentum-bias unification with RESULT 22 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-15RESULT 25: bias decomposition probe — 100% between-block transmission ↵Yuren Hao
(TRANS=EP 0.9987, ANCHOR=1.0000, K1=BP exactly, K3=K8 saturated) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-15RESULT 23 verdict: gap tracks FIT DEPTH not width — all registered ↵Yuren Hao
hypotheses falsified (C128 zero-gap at cos 0.805); fw72m_cent live at 1.95 it/s Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-14RESULT 24: fw72m_cent pre-registration (centered crown rerun, bcap-v3 0.9, ↵Yuren Hao
gradtest 0.9999877 gate passed, chained behind ladder) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-14RESULT 22 arms7 verdict (centered@3e-3 wins tail-SNR, momentum refuted) + ↵Yuren Hao
RESULT 23 gap-scaling pre-registration; ladder launched GPU1+GPU3 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-14RESULT 21: crown sealed — 72.11M x 1.44B tokens fully BP-free (largest to ↵Yuren Hao
date), gen gate passed on FineWeb; casc_gen data-aware Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-14RESULT 20: wall-zone verdicts — damping refuted (3 witnesses, monotone ↵Yuren Hao
spectrum), beta-down sails, bcap v1 starves (meter noise floor); crown completion + arms7 launched Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-14RESULT 19: wall-2 loop-gain theory, beta window formalism, probe ↵Yuren Hao
mid-verdicts, three new controllers Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-14RESULT 18: wall-2 returns at 72M/FineWeb 195k (drift crossing, beta-stress ↵Yuren Hao
confirmed by f3e3); runs killed, geta fix probes launched from s195000 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-13fw72m crown run launched (NCCL DDP prod), exact resume (opt state), RESULT ↵Yuren Hao
17, shuffle+prep scripts Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-13RESULT 16: FineWeb-Edu pipeline live + smoke passed (CE 10.5->5.8, cos ↵Yuren Hao
0.9995, zero skips) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-12stage1b_amp epoch confirm SEALED: 1.2868 vs fp32 1.2808, zero guards, 1.56x heldYuren Hao
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-12RESULT 15 (amp lossless 1.56x) + COST_MODEL v2.1 (sourced July-2026 prices, ↵Yuren Hao
EF unit fix) + Draft 6 cost/scale rewrite Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-12E-tier wave-1 ledger (RESULT 14 + HW map rows) + cost model v2 recalibrated ↵Yuren Hao
on measured H200 datapoint Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-12RESULT 13: bsign neutral at 4k (1.7382 vs 1.7314); 42M gap-chasing closed at ↵Yuren Hao
noise floor; pivot to Stage-2 pipeline + E-tier
2026-07-11RESULT 12: f3e3cont negative (1.2883 vs 1.2808) — residual gap not ↵Yuren Hao
beta-liftable; next lever = random-sign beta (single-phase, KHS-validated), then centered
2026-07-11STANDING: looped line abandoned (user directive) — cascade-only default; ↵Yuren Hao
looped = historical data for dynamics paper
2026-07-11RESULT 11: stage1b sealed (gap 0.050); beta ceiling unfound (3e-3 -> cos ↵Yuren Hao
0.9974); K exonerated on OLMo2; bf16 naive-cast dead at any beta (TF32 stays cost baseline); f3e3 CE-endpoint cont launched Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-10Queue: double-sided nudge after stage1b (0.22 = differential-measurement SNR ↵Yuren Hao
tax, escalation ladder recorded)
2026-07-10RESULT 10: epoch endpoints (BP 1.2509/1.2750 vs EP 1.4802, +0.22 erosion ↵Yuren Hao
cost); GENERATION GATE PASSED (casc_gen.py, coherent stories, no-backprop 42.75M); stage1b improved-recipe pair launched (Muon cosine via build_hybrid) Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-10cos<0.96: flagship stays as control; bf1e3 continuation from s40000 queued ↵Yuren Hao
on farm -> endpoint quad EP-ctl/EP-lift/BPx2
2026-07-10Sustained dose-response sealed (3e-4/5e-4/1e-3 -> 0.975/0.983/0.990); ship ↵Yuren Hao
--bf_late floor ramp; next-run recipe locked (OLMo2+Muon+late-ramp)
2026-07-10RESULT 8 sealed (Muon 3v3: BP 1.7098 / EP 1.7314); beta-floor dose-response ↵Yuren Hao
@25k: 3e-4/5e-4/1e-3 -> cos 0.984/0.989/0.994, stable at 1e-3; Stage-2 recipe = late floor ramp
2026-07-10Watch: late-epoch cos erosion = intrinsic late-training SNR decline; let it ↵Yuren Hao
run, BP twins are the measurement; floor-probe queued
2026-07-10RESULT 8 final: Muon generic (+0.13 both columns; BP+Muon 1.7020 vs EP+Muon ↵Yuren Hao
1.7254); Muon default from Stage-2; 3v3 fill launched
2026-07-10RESULT 9: wall-2 ELIMINATED by OLMo2 arch — epoch cleared 10-14k with zero ↵Yuren Hao
guard events (old arch: blowup@12100); cos-erosion watch armed
2026-07-10RESULT 8 prelim: Muon+beta_floor on EP = 1.7316 (-0.098 vs AdamW, n=1, zero ↵Yuren Hao
skips); retraction closed; BP+Muon control + s2 launched
2026-07-10RESULT 7: OLMo2 parity PASS (BP 1.8335 vs EP 1.8643, gap +0.031 within ↵Yuren Hao
metric noise); arch worth ~0.08 CE both columns; epoch auto-launched; Pascal unbanned (canaries green), farm running 2xBP epochs + Muon retest
2026-07-10RESULT 6 addendum: K8 also blew at 12600 -> wall-2 is true contractivity ↵Yuren Hao
crossing (K delays only); kretry demoted to mitigation; defense = OLMo2 arch > lr channel > damped-fb/jacreg
2026-07-10RESULT 6: wall-2 = EP-specific marginal under-convergence at the ↵Yuren Hao
contractivity edge (A skips16/B skips2/C skips1/D-BP clean); ship --kretry (retry marginal batches at K=kmax); launch OLMo2 parity matrix with parity-gated epoch autolaunch Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-10AUDIT: retract confounded Muon verdict; tone down parity claims (n=3, ↵Yuren Hao
best-of-noisy-val); scope depth-tax claim to 4k horizon; guard-split skip telemetry (skd/skg); add BP control arm diag_D_bp (+--resume in BP trainer); record resume confounds Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-10Stage-1 epoch blowup @12100: sig story REFUTED (only +8%, cos fine till ↵Yuren Hao
after); leading indicator=drift-guard skips -> contractivity bifurcation in nudged relaxation (cascade Hopf wall); add resume/sig0/final_ln; A/B/C diagnostic launched
2026-07-10RESULT 4: QK-norm validated parity-preserving (EP 1.8868 <= BP 1.9066); ↵Yuren Hao
qk_norm complements but does NOT replace beta-floor (sig_tok still grows); Stage 1 TinyStories epoch launched
2026-07-10QK-norm: replace nn.MHA with explicit SDPA attn + --qk_norm (OLMo2-style, ↵Yuren Hao
analog-friendly); cancel epoch, insert 8-run qk validation, stage roadmap TinyStories-epoch->FineWeb-Edu->OLMo2
2026-07-10RESULT 3: K4 depth-parity SEALED EP-favorable (EP 1.8907 <= BP 1.9192 @ ↵Yuren Hao
L12xC512); full-epoch generation run auto-launched (6.9h)
2026-07-10CORRECTION: D1a did not die (misread mangled proc scan); originals completed ↵Yuren Hao
s1 1.9745/s2 2.0013/s3 2.1188 + muon 2.7515; beta-floor fix independently found on free GPU1 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-10RESULT 2: beta-floor confirmed as L12 depth fix (3e-4 holds cos=1.0 vs ↵Yuren Hao
control collapse 0.896); K4 3-seed verdict launched vs BP 1.9192 Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-10K refuted as L12 lever (cos K-invariant K3==K8); BP 3-seed sealed 1.919; ↵Yuren Hao
mechanism = finite-beta SNR collapse; add --beta_floor/--beta_fixed + launch floor sweep Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-09D1a autopsy + K-ladder diagnostic: parent-death (not nohup'd) killed 4 arms; ↵Yuren Hao
L12 cos-erosion signal; relaunched K3/K8/BPs3 nohup Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-09notes refresh: plan status stamp (K1-K3 sealed, D-tier in flight), ↵Yuren Hao
ONBOARDING cascade-pivot section Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-09cascade: retire zil per user directive (EP route = equilibrium two-phase); ↵Yuren Hao
probe gains --init_sweep (state warm-start, EP-clean readout) and --sopt adam; B2 solver sweep Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
2026-07-09cascade: zil interleaved scheme = exact gradients at all depths (cos 1.0000 ↵Yuren Hao
L6-24, trajectory 0.9998+); casc_ep_train zil trainer; naive-relaxation depth failure documented Co-Authored-By: Claude Fable 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
generated by cgit v1.2.3 (git 2.25.1) at 2026-08-18 21:49:15 +0000