xysheng-colossal
|
84723e8bed
|
[feat][merge] Support one-behind to reduce bubble time. Add profiling code. (#6355)
* [feat][merge] Support one-behind to reduce bubble time. Add profiling code.
* [feat] Update sync model by tensor, fix tMbs problem, add qwen train benchmark.
* [feat] Update consumer init to run 32B , update qwen benchmark.
|
2025-09-02 17:05:15 +08:00 |
|