marsggbo

There are some parameters I don't know what their meanings. Sucha as: micro_child.py - sync_replicas - num_aggregate - num_replicas Are these parameters used for multi gpus?

./TensorNVMe/include/aio.h:3:10: fatal error: libaio.h: No such file or directory

3

how to solve this problem without root priority when installing tensornvme?


- https://github.com/kornia/kornia - https://pytorch-lightning.readthedocs.io/en/latest/notebooks/lightning_examples/augmentation_kornia.html

Any plan to release the search code (MCT)?

1

Any plan to release the search code for [Prioritized Architecture Sampling with Monto-Carlo Tree Search](https://arxiv.org/abs/2103.11922) (MCT)? Thanks

Do training and inference of MoE share the same dispatching method?

1

While MoE training typically uses a fixed capacity to distribute tokens evenly across all experts, my understanding is that inference involves activating experts based on predicted relevance via a softmax...

why does lightning still not support Pipeline parallelism?

3

### Description & Motivation I want to benchmark different parallelisms, but I didn't find pipeline mode. ### Pitch _No response_ ### Alternatives _No response_ ### Additional context _No response_ cc...