-
1329bb9506
Support stage1-2 MoE resume on g0050
main
yi_lu
2026-07-16 09:23:10 +08:00
-
07e108388d
Add heldout HF evaluation workflow
yi_lu
2026-07-09 08:39:02 +08:00
-
e1e880cadf
Add agent operating guide
yi_lu
2026-07-09 08:16:37 +08:00
-
2b0588ad17
Sync g0050 training and inference tooling
yi_lu
2026-07-09 08:02:25 +08:00
-
0e8c2a554d
Support g0050 SSD preprocessing path
yi_lu
2026-07-02 21:59:02 +08:00
-
b8b5c3ffe0
Use vLLM triton kernels for MoE backend
yi_lu
2026-07-02 21:29:10 +08:00
-
323452b9bd
Add g0050 NeMo flash-attn4 training image
yi_lu
2026-07-02 21:10:58 +08:00
-
5698a9cc21
Add ModelScope parquet upload helper
yi_lu
2026-07-02 20:55:21 +08:00
-
5609b1f8e4
Add Megatron data manifest and g0050 setup
yi_lu
2026-07-02 20:50:24 +08:00
-
816eccb5b5
Tie embeddings for small MoE config
yi_lu
2026-07-01 23:27:56 +08:00
-
2c2b7ccc24
Add Megatron-Bridge pretrain launcher
yi_lu
2026-07-01 23:10:37 +08:00
-
f0f72767f5
Add proof-pile logic inputs to pretrain builder
yi_lu
2026-07-01 22:42:18 +08:00
-
cdd8e2f85b
Add NeMo backend probe and architecture validator
yi_lu
2026-07-01 20:46:45 +08:00
-
11308f95c4
Use verified GLM-5.2 tokenizer from ModelScope
yi_lu
2026-07-01 19:48:17 +08:00
-
e7ebb30e1f
Use GLM-5.2 tokenizer vocabulary
yi_lu
2026-07-01 19:35:57 +08:00
-
dcabce8c07
Initialize rebalanced 2B MoE project
yi_lu
2026-07-01 16:41:54 +08:00