diff options
| author | Yuren Hao <yurenh2@illinois.edu> | 2026-07-06 07:15:30 -0500 |
|---|---|---|
| committer | Yuren Hao <yurenh2@illinois.edu> | 2026-07-06 07:15:30 -0500 |
| commit | 488c50e1bdbf8f420b2ad5b4021a7f950d121835 (patch) | |
| tree | 3605b3c92af8fd7c0c18ab5982bee084a858b015 /ep_run/runs/bp_sweep.sh | |
| parent | 8f43de671cf03ee20d5d652b6cd3ba575982a436 (diff) | |
--holoavg shipped (gate 0.913->0.936 @t2sel160, batch2 +0.051) + farm wave-2 tooling
trend-aware stop + plateau averaging recovers the semi-convergence victims
exactly as predicted. Estimator pack now: holofast+sdpa+t2sel80(+holoavg).
Also: bp_lm stdinit/beta2/sched flags (anchor archaeology), dipfarm_freezer,
staged tol_sweep.sh (gated on hr2 verdict), bp_sweep.sh.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn
Diffstat (limited to 'ep_run/runs/bp_sweep.sh')
| -rwxr-xr-x | ep_run/runs/bp_sweep.sh | 8 |
1 files changed, 8 insertions, 0 deletions
diff --git a/ep_run/runs/bp_sweep.sh b/ep_run/runs/bp_sweep.sh new file mode 100755 index 0000000..e98c79c --- /dev/null +++ b/ep_run/runs/bp_sweep.sh @@ -0,0 +1,8 @@ +#!/bin/bash +cd /home/yurenh2/ept/ep_run +for cfg in "qk3 --qknorm --lr 3e-3" "qk1 --qknorm --lr 1e-3" "nq3 --lr 3e-3" "nq1 --lr 1e-3"; do + set -- $cfg; name=$1; shift + CUDA_VISIBLE_DEVICES=0 python3 bp_lm.py "$@" --steps 32000 --log 1000 --ckpt runs/bp_$name.pt > runs/bp_$name.log 2>&1 + grep "DONE" runs/bp_$name.log +done +echo BP_SWEEP_ALL_DONE |