v0.5.5rc3-rocm
siddharthvaddem/openscreenv0.5.5rc3-rocmSep 1, 2026by github-actions[bot]
AI Summary
This release candidate provides a standard ROCm 7.2 wheel for LMCache v0.5.5rc3, optimized for AMD Instinct gfx942 and gfx950 GPUs and ABI-matched to the upstream `vllm/vllm-openai-rocm` image.
Key Highlights
- ROCm 7.2 wheel for LMCache v0.5.5rc3
- Optimized for AMD Instinct gfx942 (MI300X/MI325X) and gfx950 (MI350X/MI355X)
- ABI-matched to `vllm/vllm-openai-rocm` image
- Includes installation instructions for upstream vLLM ROCm containers
New Features
- ROCm wheel for specific AMD GPU generations
Full Release Notes
ROCm 7.2 wheel for LMCache v0.5.5rc3, built for AMD
Instinct gfx942 (MI300X/MI325X) and gfx950 (MI350X/MI355X). ABI-matched
to the upstream `vllm/vllm-openai-rocm` image (torch 2.11, cp312).
Install into an upstream vLLM ROCm container:
```
VERSION=v0.5.5rc3
pip install lmcache==0.5.5rc3+rocm7.2 --no-deps \
--find-links https://github.com/LMCache/LMCache/releases/expanded_assets/${VERSION}-rocm
```