From ad4622e42abf2a55a37c5e522a6f8302e2e95cfd Mon Sep 17 00:00:00 2001 From: Oscar Wan Date: Tue, 21 Jul 2026 15:28:06 -0700 Subject: Add portable 135M BP sweep handoff Match the active EP configuration and support four-GPU BP candidates from collaborators' data locations. Co-authored-by: Cursor --- ep_run/fw135m_baseline.py | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'ep_run/fw135m_baseline.py') diff --git a/ep_run/fw135m_baseline.py b/ep_run/fw135m_baseline.py index ddcc5c0..8daf145 100644 --- a/ep_run/fw135m_baseline.py +++ b/ep_run/fw135m_baseline.py @@ -13,7 +13,7 @@ HEADS = 12 CONTEXT = 256 BATCH = 24 PARAMETERS = 135303936 -FULL_STEPS_ARG = 440442 # casc_bp_train.py loops inclusively: 440,443 updates +FULL_STEPS_ARG = 440000 # Match fw135m_bsign; casc_bp_train.py loops inclusively: 440,001 updates. LR_SWEEP = ("7e-4", "1e-3", "1.4e-3") @@ -94,12 +94,12 @@ def main(): f"fw135m_bp_lr{lr.replace('-', 'm').replace('.', 'p')}_s{args.seed}", lr, args.sweep_steps, - 500, + 1000, ) for lr in LR_SWEEP ] else: - runs = [(f"fw135m_bp_s{args.seed}", args.lr, FULL_STEPS_ARG, 500)] + runs = [(f"fw135m_bp_s{args.seed}", args.lr, FULL_STEPS_ARG, 1000)] print( f"# L{LAYERS} C{WIDTH} H{HEADS} T{CONTEXT} B{BATCH} " -- cgit v1.2.3