biondizzle/vllm-with-lmcache
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

patch to fix MLA multi-group KV cache

Browse Source
This commit is contained in:
Matthew Harris
2026-05-06 15:48:44 -04:00
parent e286cc5208
commit d64324e233
2 changed files with 27 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

25
Dockerfile
View File

@@ -1,3 +1,26 @@
FROM vllm/vllm-openai-rocm:nightly
RUN pip install av soundfile
ENV MAX_JOBS=2
# LMCache for KV cache offloading / sharing across nodes
# Build HIP extensions for MI300X (gfx942)
RUN apt-get update && apt-get install -y git && \
git clone https://github.com/Byteflux/LMCache.git /tmp/lmcache && \
cd /tmp/lmcache && \
git checkout mla-multi-group-kv-cache-with-redis && \
pip install --no-cache-dir -r requirements/build.txt && \
BUILD_WITH_HIP=1 \
CXX=hipcc \
PYTORCH_ROCM_ARCH="gfx942" \
pip install --no-cache-dir --no-build-isolation . --verbose && \
rm -rf /tmp/lmcache && export CACHE_BUSTER=1
# Nemotron reasoning parser
COPY ./super_v3_reasoning_parser.py /opt/super_v3_reasoning_parser.py
# DeepSeek tool call parser with MTP fixes
COPY deepseekv32_tool_parser.py /usr/local/lib/python3.12/dist-packages/vllm/tool_parsers/deepseekv32_tool_parser.py
# MiniMax tool call parser with kwargs fixes
COPY minimax_tool_parser.py /usr/local/lib/python3.12/dist-packages/vllm/tool_parsers/minimax_tool_parser.py
COPY minimax_m2_parser.py /usr/local/lib/python3.12/dist-packages/vllm/parser/minimax_m2_parser.py

6
Jenkinsfile vendored
View File

@@ -7,10 +7,10 @@ pipeline {
}
parameters {
string(name: 'IMAGE_TAG', defaultValue: 'v0.19.0-cu130', description: 'Docker image tag')
string(name: 'IMAGE_TAG', defaultValue: 'rocm-mla-multi-group-kv-cache', description: 'Docker image tag')
string(name: 'GIT_REPO', defaultValue: 'https://sweetapi.com/biondizzle/vllm-with-lmcache.git', description: 'Git repository URL (optional, uses workspace if empty)')
string(name: 'GIT_BRANCH', defaultValue: 'master', description: 'Git branch to build')
string(name: 'BASE_IMAGE', defaultValue: 'vllm/vllm-openai:v0.19.0-cu130', description: 'Base Docker image')
string(name: 'GIT_BRANCH', defaultValue: 'rocm-mla-multi-group-kv-cache', description: 'Git branch to build')
string(name: 'BASE_IMAGE', defaultValue: 'vllm/vllm-openai-rocm:nightly', description: 'Base Docker image')
}
stages {