Pull request/1065 (#1068)

* Remove hugectr inference and try other test runners

* Add hps_torch installation in merlin-pytorch

* Remove the uncompatable gmock and gtest installed by triton

---------

Co-authored-by: qqiao <[email protected]>

Author: jperez999

docker/dockerfile.ctr

@@ -78,26 +78,6 @@ RUN if [[ "${HUGECTR_DEV_MODE}" == "false" ]]; then \
     ; fi
 
 ENV PYTHONPATH=${PYTHONPATH}:${HUGECTR_HOME}/lib
-ARG _HUGECTR_BACKEND_REPO="github.com/triton-inference-server/hugectr_backend.git"
-ARG TRITON_VERSION
-# Install Triton inference backend.
-RUN if [ "${HUGECTR_DEV_MODE}" == "false" ]; then \
-        git clone --branch ${HUGECTR_BACKEND_VER} --depth 1 https://${_CI_JOB_TOKEN}${_HUGECTR_BACKEND_REPO} /repos/hugectr_triton_backend && \
-        mkdir /repos/hugectr_triton_backend/build && \
-        cd /repos/hugectr_triton_backend/build && \
-        cmake \
-            -DCMAKE_INSTALL_PREFIX:PATH=${HUGECTR_HOME} \
-            -DTRITON_COMMON_REPO_TAG="r${TRITON_VERSION}" \
-            -DTRITON_CORE_REPO_TAG="r${TRITON_VERSION}" \
-            -DTRITON_BACKEND_REPO_TAG="r${TRITON_VERSION}" .. && \
-        make -j$(nproc) && \
-        make install && \
-        cd ../.. && \
-        rm -rf hugectr_triton_backend && \
-        chmod +x ${HUGECTR_HOME}/lib/*.so ${HUGECTR_HOME}/backends/hugectr/*.so && \
-        rm -rf /repos \
-    ; fi
-RUN ln -s ${HUGECTR_HOME}/backends/hugectr /opt/tritonserver/backends/hugectr
 
 # Clean up
 RUN rm -rf /usr/local/share/jupyter/lab/staging/node_modules/marked

docker/dockerfile.torch

@@ -38,6 +38,24 @@ COPY --chown=1000:1000 --from=dlfw /usr/local/lib/python${PYTHON_VERSION}/dist-p
 COPY --chown=1000:1000 --from=dlfw /usr/local/lib/python${PYTHON_VERSION}/dist-packages/numpy-*.dist-info /usr/local/lib/python${PYTHON_VERSION}/dist-packages/numpy.dist-info/
 COPY --chown=1000:1000 --from=dlfw /usr/local/lib/python${PYTHON_VERSION}/dist-packages/torch-*.egg-info /usr/local/lib/python${PYTHON_VERSION}/dist-packages/torch.egg-info/
 
+# Argumeints "_XXXX" are only valid when $HUGECTR_DEV_MODE==false
+# Install hps_torch in merlin-pytorch
+ARG HUGECTR_DEV_MODE=false
+ARG _HUGECTR_REPO="github.com/NVIDIA-Merlin/HugeCTR.git"
+ARG _CI_JOB_TOKEN=""
+ARG HUGECTR_VER=main
+
+RUN if [ "$HUGECTR_DEV_MODE" == "false" ]; then \
+	export HUGECTR_HOME=/usr/local/hugectr && \
+        rm -rf ${HUGECTR_HOME}/lib/libgmock* ${HUGECTR_HOME}/lib/pkgconfig/gmock* ${HUGECTR_HOME}/include/gmock && \
+        rm -rf ${HUGECTR_HOME}/lib/libgtest* ${HUGECTR_HOME}/lib/pkgconfig/gtest* ${HUGECTR_HOME}/include/gtest && \
+        git clone --branch ${HUGECTR_VER} --depth 1 --recurse-submodules --shallow-submodules https://${_CI_JOB_TOKEN}${_HUGECTR_REPO} /hugectr && \
+        pushd /hugectr/hps_torch/ && \
+        pip --no-cache-dir install ninja && \
+        TORCH_CUDA_ARCH_LIST="7.0 7.5 8.0 9.0" python setup.py install && \
+        popd && \
+        rm -rf /hugectr \
+    ; fi
 
 # Add all torch libraries to /usr/local
 RUN ln -s /opt/tritonserver/backends/pytorch/* /usr/local/lib/