Xin Yao comments

Repositories
Issues
Comments

Results 123 comments of


                                            Xin Yao

Avoid calling set_save_original_input with FP8 delayed scaling

/ok to test 3465f3f

Avoid calling set_save_original_input with FP8 delayed scaling

/ok to test a5f5cd5

Support Qwen3 80b Next (hybrid attn)

We already have it in the [dev branch](https://github.com/NVIDIA/Megatron-LM/tree/dev). See this [commit](https://github.com/NVIDIA/Megatron-LM/commit/c7590d8c3733619efa87a1a0733ac4cceedc683a). For the main branch, the [PR 1989](https://github.com/NVIDIA/Megatron-LM/pull/1989) is under review.