From bb2e6c5dc60c786de516619544af1085e2b61cd3 Mon Sep 17 00:00:00 2001
From: Tongze Wang <wtongze@gmail.com>
Date: Wed, 11 Sep 2024 11:23:42 -0700
Subject: [PATCH] fix: build config and depenency

---
 pytorch.dockerfile    | 30 +++++++++++++++++++++++++++---
 tensorflow.dockerfile |  7 ++++---
 2 files changed, 31 insertions(+), 6 deletions(-)

diff --git a/pytorch.dockerfile b/pytorch.dockerfile
index 6efa626..8e1b95b 100644
--- a/pytorch.dockerfile
+++ b/pytorch.dockerfile
@@ -2,7 +2,9 @@ ARG ROCM_VERSION="5.7.3"
 
 FROM wtongze/rocm-base:${ROCM_VERSION}
 
-ARG PYTORCH_BRANCH="v2.4.1"
+ARG ROCM_VERSION="5.7.3"
+ARG PYTORCH_BRANCH="v2.2.2"
+ARG PYTORCH_BUILD_VERSION="2.2.2"
 ARG PYTHON_VERSION="3.11"
 ARG HIP_ARCH="gfx803;gfx900"
 ARG DEBIAN_FRONTEND="noninteractive"
@@ -27,11 +29,33 @@ EOF
 
 RUN python${PYTHON_VERSION} -m pip install typing-extensions wheel
 
+# For ROCm 4.5+
+RUN <<EOF
+sudo apt-get update
+sudo apt-get install -y rocm-ml-sdk rocm-hip-sdk
+sudo apt-get clean
+sudo rm -rf /var/lib/apt/lists/*
+EOF
+
+RUN <<EOF
+sudo apt-get update
+sudo apt-get install -y ninja-build
+sudo apt-get clean
+sudo rm -rf /var/lib/apt/lists/*
+EOF
+
+ENV ROCM_PATH=/opt/rocm-${ROCM_VERSION}
 ENV PYTORCH_ROCM_ARCH=${HIP_ARCH}
-RUN python${PYTHON_VERSION} tools/amd_build/build_amd.py && USE_ROCM=1 python${PYTHON_VERSION} setup.py bdist_wheel
+ENV PYTORCH_BUILD_VERSION=${PYTORCH_BUILD_VERSION}
+ENV PYTORCH_BUILD_NUMBER=0
+RUN python${PYTHON_VERSION} tools/amd_build/build_amd.py && \
+USE_ROCM=ON USE_FLASH_ATTENTION=OFF USE_MEM_EFF_ATTENTION=OFF python${PYTHON_VERSION} setup.py bdist_wheel
+
+# https://github.com/pytorch/pytorch/issues/125230
+# https://github.com/AngryLoki/pytorch/commit/38d6f115c65b67946de6765165ece0288c8416d4
 
 WORKDIR /home/rocm/pytorch/dist
-RUN cp (ls -l *.whl | awk '{ print $9 }' | tail -n1 | tr -d '\n') /home/rocm/pytorch-rocm.whl
+RUN cp $(ls -l *.whl | awk '{ print $9 }' | tail -n1 | tr -d '\n') /home/rocm/pytorch-rocm.whl
 
 WORKDIR /home/rocm
 CMD [ "bash", "-l" ]
diff --git a/tensorflow.dockerfile b/tensorflow.dockerfile
index acca18f..ffcbf7f 100644
--- a/tensorflow.dockerfile
+++ b/tensorflow.dockerfile
@@ -2,6 +2,7 @@ ARG ROCM_VERSION="5.7.3"
 
 FROM wtongze/rocm-base:${ROCM_VERSION}
 
+ARG ROCM_VERSION="5.7.3"
 ARG TENSORFLOW_BRANCH="r2.12-rocm-enhanced"
 ARG PYTHON_VERSION="3.11"
 ARG NUMPY_VERSION="1.24.3"
@@ -47,11 +48,11 @@ RUN sed -i -e "s/python3/python${PYTHON_VERSION}/g" -e "s/pip3/python${PYTHON_VE
 RUN sudo ln -s /usr/bin/python${PYTHON_VERSION} /usr/bin/python
 
 ENV ROCM_PATH=/opt/rocm-${ROCM_VERSION}
-ENV ROCM_TOOLKIT_PATH=/opt/rocm-${ROCM_VERSION}
-RUN ./build_rocm_python3 /opt/rocm-${ROCM_VERSION}
+ENV ROCM_TOOLKIT_PATH=${ROCM_PATH}
+RUN ./build_rocm_python3 ${ROCM_PATH}
 
 WORKDIR /tmp/tensorflow_pkg
-RUN cp (ls -l tensorflow*.whl | awk '{ print $9 }' | tail -n1 | tr -d '\n') /home/rocm/tensorflow-rocm.whl
+RUN cp $(ls -l tensorflow*.whl | awk '{ print $9 }' | tail -n1 | tr -d '\n') /home/rocm/tensorflow-rocm.whl
 
 WORKDIR /home/rocm
 CMD [ "/bin/bash", "-l" ]