Kazuki Fujii

icon company Institute of Science Tokyo icon location Tokyo Japan icon location Master (Computer Science) student at Tokyo Institute of Technology

Results 2 issues of Kazuki Fujii

[BUG] distributed optimizer doesn't work when data parallel size is odd number.

2 comment

**Describe the bug** When the data parallel size is odd while distributed optimizer is enabled, training stops with the following error. ```bash [Proxy Service 0] Failed to execute operation Connect...

Fix: checkpoint load bug (TE)

1 comment

## Issue When using TransformerEngine with Megatron-LM for training, I encountered an issue where the Loss Curve would significantly change after loading a checkpoint. This problem did not occur when...

stale