diff options
| author | Yuren Hao <yurenh2@illinois.edu> | 2026-07-06 09:42:52 -0500 |
|---|---|---|
| committer | Yuren Hao <yurenh2@illinois.edu> | 2026-07-06 09:42:52 -0500 |
| commit | 40be67d4f5b5a6b46c662c70b759e585e136d70e (patch) | |
| tree | bbe02d5164c9b86a0af6a3634fb7556f065c8672 /ep_run/fastfp_gate.log | |
| parent | 9a8b2796ca12e4d4c24717485a635a301aa6d07f (diff) | |
Tier-3 gates: Anderson math-yes/impl-no (parked for v2); bf16polish UNSAFE near-edge (eval-only); dp_ep.py ready
Anderson: res 25-35x deeper per budget but 5.8x slower (naive history stacks
+ per-iter safeguard eval) — v2 = ring buffers + periodic safeguard, est +1.3x
on the speed tier. bf16+20polish: 1.41x free phase, res parity, BUT z-diff
1.2e-3 — near-marginal operators contract too slowly for a 20-step polish
(0.998^20≈0.96), same magnitude as the TF32 kill verdict and the estimator's
50%-sensitivity input. Predicted by our own depth/noise theory. Flags kept
with warnings; neither ships for training. dp_ep.py: manual-allreduce EP DP
(controller in lockstep, aligned collectives), smoke pending freed 1080s.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
Diffstat (limited to 'ep_run/fastfp_gate.log')
| -rw-r--r-- | ep_run/fastfp_gate.log | 7 |
1 files changed, 7 insertions, 0 deletions
diff --git a/ep_run/fastfp_gate.log b/ep_run/fastfp_gate.log new file mode 100644 index 0000000..5beee9a --- /dev/null +++ b/ep_run/fastfp_gate.log @@ -0,0 +1,7 @@ +[s2000] euler150 : 1.85s res=9.98e+00 +[s2000] anderson : 10.75s res=3.88e-01 evals=150 z-diff=5.01e-02 +[s2000] bf16+20 : 1.31s res=9.96e+00 z-diff=1.19e-03 +[fast-adaptive] euler150 : 1.64s res=3.31e+00 +[fast-adaptive] anderson : 10.54s res=9.49e-02 evals=150 z-diff=3.81e-02 +[fast-adaptive] bf16+20 : 1.21s res=3.33e+00 z-diff=1.13e-03 +FASTFP_GATE_DONE |