mirror of
https://github.com/hanzo-ml/kserve.git
synced 2026-10-30 20:53:49 +00:00
26348e4a48
Signed-off-by: cory-johannsen <[email protected]> Signed-off-by: Cory Johannsen <[email protected]> Co-authored-by: Sivanantham <[email protected]>
33 lines
1.2 KiB
Makefile
33 lines
1.2 KiB
Makefile
IPEX_EXTRA_INDEX_URL = https://pytorch-extension.intel.com/release-whl/stable/cpu/us/
|
|
TORCH_EXTRA_INDEX_URL = https://download.pytorch.org/whl/cpu
|
|
|
|
dev_install:
|
|
uv sync --active --group test
|
|
|
|
install_dependencies:
|
|
uv sync --active --group test
|
|
|
|
# CPU build workflow:
|
|
# 1. First run install_cpu_dependencies to install huggingfaceserver dependencies
|
|
# 2. Then run setup_vllm.sh LAST to build vLLM from source with torch CPU
|
|
#
|
|
# kserve[llm] pulls vllm from PyPI as a pre-built GPU wheel, which brings in GPU
|
|
# torch. The CPU build instead needs vllm built from source with torch+cpu. These
|
|
# two installation paths are incompatible, so we temporarily strip the [llm] extra
|
|
# to prevent PyPI vllm (and GPU torch) from being installed. setup_vllm.sh then
|
|
# handles the entire vllm + torch CPU installation from source.
|
|
install_cpu_dependencies:
|
|
@echo "Replacing kserve[llm] with kserve to avoid vllm dependency conflict..."
|
|
sed -i 's/kserve\[llm\]/kserve/' pyproject.toml
|
|
uv lock
|
|
@echo "Installing huggingfaceserver with test dependencies..."
|
|
uv sync --active --group test
|
|
@echo "Restoring original pyproject.toml..."
|
|
sed -i 's/"kserve @/"kserve[llm] @/' pyproject.toml
|
|
|
|
test: type_check
|
|
pytest -W ignore
|
|
|
|
type_check:
|
|
mypy --ignore-missing-imports huggingfaceserver
|