build-rocm-image

To build a ROCm Docker image on a remote host, you need SSH access and a scripted workflow that generates a remote Dockerfile from the rocm/vllm-dev:nightly base image, clones required components, and performs the build.

Yes, you can install FlyDSL and rocprofv3 together by automating a Docker build workflow that clones FlyDSL from the develop branch and consolidates it with rocprofv3 and aiter into a single reproducible remote ROCm dev environment.

Setting up a remote ROCm development environment requires SSH access to the target host and the rocm/vllm-dev:nightly base image to serve as the foundation for the scripted Docker build process.

You verify installed versions and branches after building a ROCm dev image by running scripted validation steps that report the specific branch info for components like triton and vllm, ensuring the reproducible build succeeded.

Building this ROCm dev image requires a custom triton cloned specifically from the rocm-maxnreg-support-v35 branch, alongside vllm from the ps_pa branch, to ensure precise component versioning.

The best way to run a ROCm container with GPU access after a remote build is to follow the provided post-build instructions, which validate the image and detail the specific Docker run commands needed for GPU access.