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

Compare commits

merge into: biondizzle:dream-build
Branches Tags
biondizzle:master biondizzle:dream-build-glm biondizzle:dream-build biondizzle:rocm-mla-multi-group-kv-cache biondizzle:dream-build-glm-save-decode biondizzle:mla-multi-group-kv-cache biondizzle:glm51-cu130-nightly biondizzle:openai-rocm-nightly biondizzle:deepseekv4-cu130 biondizzle:custom-weights-cu130 biondizzle:glm51-cu130
...
pull from: biondizzle:rocm-mla-multi-group-kv-cache
Branches Tags
biondizzle:dream-build-glm biondizzle:dream-build biondizzle:rocm-mla-multi-group-kv-cache biondizzle:dream-build-glm-save-decode biondizzle:mla-multi-group-kv-cache biondizzle:glm51-cu130-nightly biondizzle:openai-rocm-nightly biondizzle:deepseekv4-cu130 biondizzle:custom-weights-cu130 biondizzle:master biondizzle:glm51-cu130

4 Commits
dream-buil ... rocm-mla-m

Author SHA1 Message Date
Matthew Harris
951c886071 follow-up LocalCPUBackend fix 2026-05-06 20:07:29 -04:00
Matthew Harris
d64324e233 patch to fix MLA multi-group KV cache 2026-05-06 15:48:44 -04:00
biondizzle
e286cc5208 add audio support 2026-05-05 05:12:44 +00:00
biondizzle
82330a31b1 add await to redis write 2026-04-17 07:19:15 +00:00
2 changed files with 20 additions and 29 deletions
Expand all files Collapse all files

43
Dockerfile
View File

@@ -1,35 +1,26 @@
#FROM vllm/vllm-openai:v0.19.0-cu130
FROM vllm/vllm-openai:cu130-nightly-x86_64
FROM vllm/vllm-openai-rocm:nightly
# Fix the broken ass nightly build that forgot to include pandas
RUN pip install --no-cache-dir pandas
ENV MAX_JOBS=2
# Install LMCache for KV cache offloading / sharing across nodes
# Build with system CUDA 13.0 for Blackwell (B200)
RUN apt-get update && apt-get install -y git \
libcusolver-dev-13-0 \
libcusparse-dev-13-0 \
libcublas-dev-13-0 \
libcurand-dev-13-0 \
libcufft-dev-13-0 \
libnvjitlink-dev-13-0 && \
git clone https://github.com/biondizzle/LMCache.git /tmp/lmcache && \
# 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 feat/redis-ttl && \
CUDA_HOME=/usr/local/cuda \
TORCH_CUDA_ARCH_LIST="10.0" \
pip install --no-cache-dir --no-build-isolation . && \
rm -rf /tmp/lmcache
git checkout mla-multi-group-kv-cache && \
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
# Copy over nemotron reasonong parser
# Nemotron reasoning parser
COPY ./super_v3_reasoning_parser.py /opt/super_v3_reasoning_parser.py
# Copy over deepseek tool call parser with MTP fixes
# 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
# Copy over minimax tool call parser with kwargs fixes
# 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 over minimax parsers 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
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 {