Nemo and 4090

AI & Data Science Deep Learning (Training & Inference) Riva
1

I’ve been trying to work through this tutorial for Nemo:

https://developer.nvidia.com/blog/deploying-a-1-3b-gpt-3-model-with-nvidia-nemo-megatron/

which suggested this in the System Requirements:

  • An NVIDIA Ampere architecture GPU or newer with at least 8 GB of GPU memory.

However, when I tried to run the model conversion on my 4090-equipped Ubuntu 20.04 box, I got this error:

WARNING: Detected NVIDIA NVIDIA GeForce RTX 4090 GPU, which is not yet supported in this version of the container
ERROR: No supported GPU(s) detected to run this container

The containers I pulled for training and inference had the tags 22.08.01-py3 and 22.08-py3, respectively.

$ docker pull nvcr.io/ea-bignlp/bignlp-training:22.08.01-py3
$ docker pull nvcr.io/ea-bignlp/bignlp-inference:22.08-py3

The article does suggest getting the latest tags, but I wasn’t sure how to find the latest tags.

So, my question is, how do I find the latest tags? And, is Nemo really not supported on the 4090?

Thanks!

2

Hi @rbgreenway

Thanks for your interest in Riva,

The request seems to be more inclined towards Nemo,

I will try to check internally and get answers, in meantime i highly recommend to start a discussion or post your question in this below Nemo github link

Thanks

3

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.