No module named 'nemo.collections.diffusion.mcore_parallel_utils'

[MtYCNN]

I have already installed NeMo (version 2.2.0rc0), Nemo toolkit, and so on. Why do I keep reporting "No module named 'nemo. collections. diffusion. mcore_prallel_utils'" when performing multi GPU inference?

[ethanhe42]

did you use our provided container? version 2.2.0rc0 doesn't have the supporting code yet.
or you can try this branch https://github.com/NVIDIA/NeMo/tree/physicalai/nemo/collections/diffusion

[ethanhe42]

you can try build container with this Dockerfile

FROM nvcr.io/nvidia/nemo:24.12

RUN pip install --no-cache-dir imageio[ffmpeg] pyav iopath better_profanity peft git+https://github.com/NVlabs/Pytorch_Retinaface.git@b843f45

ARG GITLAB_CLONE_ACCESS_TOKEN
RUN rm -rf /opt/megatron-lm
RUN git clone --single-branch --branch cosmos [email protected]:NVIDIA/Megatron-LM.git /opt/megatron-lm && cd /opt/megatron-lm && pip install -e . && rm -rf .git && cd megatron/core/datasets && make
RUN rm -rf /opt/NeMo
RUN git clone --single-branch --branch physicalai [email protected]:NVIDIA/NeMo.git /opt/NeMo && cd /opt/NeMo && git checkout 75aa5fe3e5acc5a885bc3409be553960067fabb9 && rm -rf .git