summaryrefslogtreecommitdiff
path: root/ep_run/casc_eq_train.py
diff options
context:
space:
mode:
authorYuren Hao <yurenh2@illinois.edu>2026-07-10 20:56:01 -0500
committerYuren Hao <yurenh2@illinois.edu>2026-07-10 20:56:01 -0500
commit6d9fcd748df0836528b31f29717ac3bc8715d1dd (patch)
tree288abf1e70f7d955f7d5d008d9648f3053986a00 /ep_run/casc_eq_train.py
parent196ee7d8121fccaae630e408bf29f6664ff351f5 (diff)
RESULT 10: epoch endpoints (BP 1.2509/1.2750 vs EP 1.4802, +0.22 erosion 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
Diffstat (limited to 'ep_run/casc_eq_train.py')
-rw-r--r--ep_run/casc_eq_train.py3
1 files changed, 2 insertions, 1 deletions
diff --git a/ep_run/casc_eq_train.py b/ep_run/casc_eq_train.py
index 6bd83f8..5a3e640 100644
--- a/ep_run/casc_eq_train.py
+++ b/ep_run/casc_eq_train.py
@@ -175,7 +175,8 @@ if args.resume:
print(f'[resume] loaded {args.resume} at step {start_step}', flush=True)
if args.opt == 'muon':
from muon import build_hybrid
- opt, sched = build_hybrid(blocks, all_params, args.lr, args.muon_lr, args.warmup)
+ opt, sched = build_hybrid(blocks, all_params, args.lr, args.muon_lr, args.warmup,
+ total_steps=(args.steps if args.cosine else 0), lr_min_ratio=args.lr_min_ratio)
else:
if args.wd >= 0: # OLMo2-style grouped decay: linear weights + head decay; embeddings/norm-gains none
nodecay = {id(p) for p in tok.parameters()} | {id(p) for p in pos.parameters()} | \