7 Commits

Author	SHA1	Message	Date
yi_lu	1329bb9506	Support stage1-2 MoE resume on g0050	2026-07-16 09:23:10 +08:00
yi_lu	2b0588ad17	Sync g0050 training and inference tooling	2026-07-09 08:02:25 +08:00
yi_lu	5609b1f8e4	Add Megatron data manifest and g0050 setup	2026-07-02 20:50:24 +08:00
yi_lu	816eccb5b5	Tie embeddings for small MoE config	2026-07-01 23:27:56 +08:00
yi_lu	2c2b7ccc24	Add Megatron-Bridge pretrain launcher	2026-07-01 23:10:37 +08:00
yi_lu	f0f72767f5	Add proof-pile logic inputs to pretrain builder	2026-07-01 22:42:18 +08:00
yi_lu	dcabce8c07	Initialize rebalanced 2B MoE project	2026-07-01 16:41:54 +08:00