xiongliangcheng 2b84f8f52a !1658 【mcore】添加baichuan2-13B和deepseek2-coder适配 ... Merge pull request !1658 from xiongliangcheng/deepseek-coder		2024-09-12 01:04:21 +00:00
..
datasets	!1615 【st提交】增加varlen场景下，FA layout=TND st用例	2024-09-05 06:44:02 +00:00
distributed	!1614 TTP/UCE高可用特性tests/README.md用例维护	2024-09-05 02:48:15 +00:00
models	!1642 mod partial_rope to glm_rope	2024-09-10 06:14:35 +00:00
optimizer	!1629 使用ms-core的reuse fp32 param实现	2024-09-10 01:05:47 +00:00
pipeline_parallel	!1329 ModelLink配套升级到megatron core 0.6.0	2024-06-11 07:53:57 +00:00
tensor_parallel	!1578 [mcore-llm]类deepseekv2模型性能优化：无tp场景下的大词表mm本地切分，并支持mla场景的指定softmax_scale	2024-08-31 01:13:32 +00:00
transformer	!1658 【mcore】添加baichuan2-13B和deepseek2-coder适配	2024-09-12 01:04:21 +00:00
__init__.py	!1585 feat: 使用更高性能的all gather permutation和 un-permutation实现	2024-09-03 01:08:31 +00:00
parallel_state.py	!1527 去除dp对ep断言，ep在dpxcp域使能	2024-08-19 11:31:39 +00:00