Open DhanshreeA opened 7 months ago
Hi @DhanshreeA
Please, it is not good practice to leave issues without explanation just open. Kindly update/rewrite or close if no longer an issue.
My bad, @GemmaTuron . I have updated the issue.
Not sure if this is important for this issue: https://github.com/NVIDIA/bionemo-framework
Summary
Several generative models within Ersilia, particularly those used by ChemSampler, would benefit with running on GPU rather than CPU to speed up performance. However, due to the fact that we want to make Ersilia available in low resource settings, we set out with all the models only supported on CPU runtime. This is presently posing as a problem especially during dogfooding ChemSampler, where despite having access to GPUs, we are unable to actually leverage its performance boost. Therefore, it makes sense to support Ersilia models both for CPU and GPU runtimes.
Objectives
This presently does not have a detailed roadmap so we cannot list clear objectives, however we need to time box and estimate the effort required to make this possible. To start off with, we will select a few models where we want to achieve GPU support and see if we can come up with a scalable solution.
Documentation
No response