Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Install PyTorch from nightly wheels #3

Merged
merged 5 commits into from
Oct 16, 2024
Merged

Install PyTorch from nightly wheels #3

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
62c4c4b
Install PyTorch from nightly wheels
rostan-t Oct 11, 2024
160f308
Remove the mention of the PyTorch target from the README
rostan-t Oct 11, 2024
1a4eaf9
Add proper quoting when checking for PyTorch wheels CUDA compatibility
rostan-t Oct 14, 2024
4a24a65
Enforce CUDA 12.4.x for PyTorch and set the default CUDA version to 1…
rostan-t Oct 14, 2024
2ee6d29
Use POSIX-compliant way to remove dots in CUDA version
rostan-t Oct 15, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
5 changes: 0 additions & 5 deletions README.rst
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,11 +29,6 @@ How to try it?

$ docker run -it --gpus all -v test:/test free-threaded-python python3 /test/simple.py

#. If you want to build the container with PyTorch, specify the target (this takes some time)::

$ docker build -t free-threaded-python --target pytorch .


This is an experimental software!
---------------------------------

Expand Down
35 changes: 9 additions & 26 deletions docker/Dockerfile
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -19,9 +19,9 @@
# FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
# DEALINGS IN THE SOFTWARE.

ARG CUDA_VERSION=12.5.0
ARG CUDA_VERSION=12.4.0

FROM nvcr.io/nvidia/cuda:${CUDA_VERSION}-devel-ubuntu22.04 as base
FROM nvcr.io/nvidia/cuda:${CUDA_VERSION}-devel-ubuntu22.04

ARG DEBIAN_FRONTEND=noninteractive
ENV PYTHON_GIL=0
Expand Down Expand Up @@ -51,6 +51,13 @@ RUN pip install numpy
# Cython and Pillow nightly builds
RUN python3 -m pip install -i https://pypi.anaconda.org/scientific-python-nightly-wheels/simple cython pillow

# PyTorch nightly build
RUN if [ "$(echo "$CUDA_VERSION" | tr -d . | head -c 3)" != 124 ]; then \
echo "No available free-threaded PyTorch wheels for CUDA version $CUDA_VERSION"; \
else \
python3 -m pip install --pre torch --index-url https://download.pytorch.org/whl/nightly/cu124; \
fi

# Install nvImageCodec from source
# Build dependencies
RUN python3 -m pip install setuptools wheel clang==14 libclang==14.0.1
Expand Down Expand Up @@ -102,27 +109,3 @@ RUN wget -O nsight.deb https://developer.nvidia.com/downloads/assets/tools/secur

RUN rm -r /opt/nvidia/entrypoint.d/*
COPY entrypoint.d /opt/nvidia/entrypoint.d


ARG TEST_PYTORCH=0
# Stage specifically for PyTorch
FROM base as pytorch

WORKDIR /opt

# Install optree from source. Without it, we get FPEs at runtime
RUN python3 -m pip install 'git+https://github.com/metaopt/optree.git#egg=optree'

# Install pytorch from source
# Clone (this takes a very long time)
RUN git clone --recursive -j"$(grep ^processor /proc/cpuinfo | wc -l)" https://github.com/pytorch/pytorch
# Patch and install
RUN cd pytorch && git remote add alband https://github.com/albanD/pytorch.git && \
{ git fetch alband || true; } && git switch update_313t && \
git submodule update --init --recursive && git -C third_party/fmt checkout 10.2.1 && \
python3 -m pip install -r requirements.txt && \
MAX_JOBS="$(grep ^processor /proc/cpuinfo | wc -l)" BUILD_TEST=$TEST_PYTORCH python3 setup.py bdist_wheel install
RUN cd pytorch && python3 -m pip install ./dist/*.whl

# Default stage - PyTorch is not built.
FROM base as default