Amrit Bulusu 641e8a6dd3 muP implementation: coord check, transfer check, and code quality fixes ... - Fix output logit hook in coord check to apply muP scaling (base/width) - Replace config mutation side effect with assertion in setup_optimizer - Set mup_base_width at GPTConfig construction in base_train.py - Remove dead code (_transfer_check_output_mult) - Tune base LRs to center optimal multiplier near 1.0 (0.12, 6.0, 0.12) - Use log scale on all loss plots for better low-loss detail - Add automated muP tests (coord check + transfer check) - Update muP_changes.md verification commands Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>		2026-03-14 16:28:50 -04:00
..
base_eval.py	delete autocast, an unnecessary thorn in my side, manage dtypes directly	2026-03-04 23:55:30 +00:00
base_train.py	muP implementation: coord check, transfer check, and code quality fixes	2026-03-14 16:28:50 -04:00
chat_cli.py	delete autocast, an unnecessary thorn in my side, manage dtypes directly	2026-03-04 23:55:30 +00:00
chat_eval.py	delete autocast, an unnecessary thorn in my side, manage dtypes directly	2026-03-04 23:55:30 +00:00
chat_rl.py	delete autocast, an unnecessary thorn in my side, manage dtypes directly	2026-03-04 23:55:30 +00:00
chat_sft.py	delete autocast, an unnecessary thorn in my side, manage dtypes directly	2026-03-04 23:55:30 +00:00
chat_web.py	delete autocast, an unnecessary thorn in my side, manage dtypes directly	2026-03-04 23:55:30 +00:00
mup_coord_check.py	muP implementation: coord check, transfer check, and code quality fixes	2026-03-14 16:28:50 -04:00
mup_transfer_check.py	muP implementation: coord check, transfer check, and code quality fixes	2026-03-14 16:28:50 -04:00
tok_eval.py	initial commit	2025-10-13 06:49:24 -07:00
tok_train.py	quick fix to not OOM main speedrun script	2026-01-26 22:31:42 +00:00