From 6d9fcd748df0836528b31f29717ac3bc8715d1dd Mon Sep 17 00:00:00 2001 From: Yuren Hao Date: Fri, 10 Jul 2026 20:56:01 -0500 Subject: 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 Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn --- ep_run/casc_eq_train.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'ep_run/casc_eq_train.py') 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()} | \ -- cgit v1.2.3