cosine decay

pr_number	parameters
435	{"warmup":true}
473	{"across_chunks":true}
481	{"epochs":30}
508	{"over_actual_training_window":true,"chunks":200}
509	—
517	{"t_max":100,"eta_min":0.00001}
518	{"epochs":50,"formula":"lr = 0.5 (1 + cos(pi * progress))"}
528	—
532	{"ttt_epochs":10}
537	{"intra_chunk":true,"inter_chunk":true,"formula":"0.5 × (1 + cos(π × step / total_steps))"}
545	{"across_chunks":true}
549	{"warmdown_steps":3500}
562	{"per_step":true,"decay_to":0}
568	{"start_lr":0.01,"end_lr":0.001,"epochs":5}
573	{"to_zero":true,"warmdown_steps":3500}
578	{"fixed_window":200}
581	{"start_lr":0.0005,"end_lr":0.00002,"warmup_epochs":1}
585	{"across_chunks":true}
589	{"learning_rate":0.002,"applied_to":"TTT across chunks"}
596	{"min_lr_fraction":0.1,"within_ttt":true}
606	—
607	—
611	{"ttt_epochs":8}
614	{"per_step":true}
642	{"across_total_ttt_steps":true}
653	—
661	{"ttt_epochs":30}
672	{"phase":"TTT","epochs":30}
685	{"epochs":20}
688	{"adaptive_lr_max_mult":3}
691	{"epochs":30}
702	{"scheduler":"CosineAnnealingLR","t_max":100,"eta_min":0.00001}
714	{"warmdown_steps":3500}
720	{"warmdown":true}
726	{"across_chunks":true}
741	—
752	{"across_chunks":true}
771	{"epochs":30,"final_lr":0}
809	{"warmdown_steps":3500}
826	—
834	{"across_chunks":true}
838	{"warmdown_steps":3500}
857	{"phase":"test-time training","epochs":20}
880	{"adaptive":true}
885	{"across_chunks":true}
914	{"warmdown_iters":3000}
953	{"within_ttt":true}
960	{"start_lr":0.008,"end_lr":0.00001}
967	{"within_ttt":true}
991	—
994	{"warmup":true,"warmdown":true}
995	—
999	{"warmdown_steps":3500}
1039	{"ttt":true}
1092	{"warmdown_steps":3500}
1127	—
1148	{"warmdown_steps":3500}
1186	—
1202	{"applied_to":"TTT chunks"}
1218	—
1227	—
1229	{"start_lr":0.008,"end_lr":0.0008,"steps":16}
1230	{"used_for":"TTT"}
1231	{"used_for_ttt":true}
1235	{"warmup_steps":2,"main_scale":0.5,"min_scale":0.05,"gamma":0.8}
1239	—
1257	—
1269	{"warmdown_steps":3500}
1280	{"warmdown_iters":3500}
1306	—
1313	{"start_lr":0.012,"end_lr":0.001}
1320	{"per_chunk":true}
1321	{"start_lr":0.012,"end_lr":0.001}
1324	{"start":0.012,"end":0.001}
1328	{"start_lr":0.024,"end_lr":0.001}
1341	{"chunk_size":32768}
1370	—
1380	{"min_lr_frac":0.1}
1388	—
1400	{"epochs":10}
1406	{"applied_to":"TTT"}
1408	{"warmdown_steps":4000}
1423	—
1430	{"lr_start":0.432,"lr_end":0.001}
1431	{"used_for":"TTT","lr":0.003}
1434	—
1457	—
1460	—
1485	—
1487	—
1488	{"start":0.012,"end":0.001}
1489	—
1492	—
1493	—
1501	{"warmdown_start_step":2200}
1502	{"phase":"warmdown"}
1517	{"ttt":true}
1532	{"warmdown":0.72}
1539	{"final_lr_factor":0.1}
1544	{"warmup_steps":100}
1545	{"warmup_steps":100}
1550	{"final_multiplier":0.1}
1553	{"warmup_steps":100,"constant_after_warmup":true}
1580	{"warmup_steps":64,"decay_start_frac":0.65,"min_scale":0.15}
1627	—
1636	{"decay_steps":3000}
1696	{"applied_to":"TTT"}
1698	{"applied_to":"TTT learning rate"}
1711	—
1714	{"warmup_steps":20,"warmdown_fraction":0.72}
1722	{"start_lr":0.024,"end_lr":0.001}
1731	{"applied_to":"TTT"}
1735	{"scope":"epoch-level","t_max":21,"eta_min_ratio":0.1}
1737	{"across_chunks":true}
1738	{"scope":"epoch-level","t_max":21,"eta_min_ratio":0.1}
1758	{"epochs":21}