feat(gpu): add AMD ROCm GPU support for local embeddings #199
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Add two AMD GPU modes: - amd: Uses ROCm PyTorch wheels (for ROCm 6.x systems) - amd-host: Uses official ROCm Docker image (for ROCm 7.x systems) The amd-host mode uses rocm/pytorch:rocm7.1 base image with: - ROCm 7.1 with full HIP/HSA support - PyTorch 2.9.1 with ROCm backend - Python 3.13 Key changes: - New docker-compose.gpu-amd.yml for ROCm wheels mode - New docker-compose.gpu-amd-host.yml for host ROCm mode - New api/Dockerfile.rocm-host using official ROCm image - Updated api/Dockerfile with PYTORCH_VARIANT build arg - Updated guided-init.sh with AMD GPU options (3 and 4) - Updated common.sh and start-app.sh for new GPU modes Fixed: Empty HSA_OVERRIDE_GFX_VERSION env var breaks ROCm initialization. Removed from compose files with documentation note that users can set in .env if needed. Tested on AMD Radeon RX 7900 XTX with ROCm 7.1.1 host driver.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary
Adds AMD GPU (ROCm) support for GPU-accelerated local embeddings, complementing existing NVIDIA CUDA and Apple MPS support.
Two AMD GPU modes:
amd: Uses PyTorch ROCm wheels (for ROCm 6.x systems)amd-host: Uses official ROCm Docker image (for ROCm 7.x systems)The
amd-hostmode usesrocm/pytorch:rocm7.1base image with:Key changes:
docker-compose.gpu-amd.ymlfor ROCm wheels modedocker-compose.gpu-amd-host.ymlfor host ROCm modeapi/Dockerfile.rocm-hostusing official ROCm imageapi/DockerfilewithPYTORCH_VARIANTbuild argguided-init.shwith AMD GPU options (3 and 4)common.shandstart-app.shfor new GPU modes.env.exampleBug fix: Empty
HSA_OVERRIDE_GFX_VERSIONenv var breaks ROCm initialization. Removed from compose files with documentation that users can set in.envif needed.Test Plan
🎮 CUDA GPU detected - AMD Radeon RX 7900 XTX (24.0GB VRAM)