pytorch · vmoens · Apr 29, 2025 · Apr 29, 2025 · Apr 29, 2025 · Apr 29, 2025
diff --git a/.github/scripts/pre-build-script-win.sh b/.github/scripts/pre-build-script-win.sh
@@ -3,7 +3,3 @@
 pip install --upgrade setuptools
 
 export TORCHRL_BUILD_VERSION=0.8.0
-
-${CONDA_RUN} pip install "pybind11[global]"
-${CONDA_RUN} conda install anaconda::cmake -y
-${CONDA_RUN} pip install git+https://github.com/pytorch/tensordict.git -U
diff --git a/.github/scripts/td_script.sh b/.github/scripts/td_script.sh
@@ -9,29 +9,20 @@ ARCH=${ARCH:-}  # This sets ARCH to an empty string if it's not defined
 if pip list | grep -q torch; then
     echo "Torch is installed."
 
-    ${CONDA_RUN} pip install "pybind11[global]"
+    # ${CONDA_RUN} conda install 'anaconda::cmake>=3.22' -y
 
-    if conda list cmake | grep -q 'cmake'; then
-        echo "CMake is already installed."
-    else
-        echo "CMake is not installed. Installing now..."
-        ${CONDA_RUN} conda install anaconda::cmake -y --no-update-deps
-    fi
+    ${CONDA_RUN} pip install "pybind11[global]"
 
     ${CONDA_RUN} pip install git+https://github.com/pytorch/tensordict.git -U --no-deps
 elif [[ -n "${SMOKE_TEST_SCRIPT:-}" ]]; then
     ${CONDA_RUN} ${PIP_INSTALL_TORCH}
     #    TODO: revert when nightlies of tensordict are fixed
     #    if [[ "$ARCH" == "aarch64" ]]; then
 
-    ${CONDA_RUN} pip install "pybind11[global]"
 
-    if conda list cmake | grep -q 'cmake'; then
-        echo "CMake is already installed."
-    else
-        echo "CMake is not installed. Installing now..."
-        ${CONDA_RUN} conda install anaconda::cmake -y --no-update-deps
-    fi
+#     ${CONDA_RUN} conda install 'anaconda::cmake>=3.22' -y
+
+     ${CONDA_RUN} pip install "pybind11[global]"
 
     ${CONDA_RUN} pip install git+https://github.com/pytorch/tensordict.git -U --no-deps
 else

diff --git a/.github/unittest/linux_libs/scripts_gym/setup_env.sh b/.github/unittest/linux_libs/scripts_gym/setup_env.sh
@@ -12,8 +12,7 @@ this_dir="$( cd "$( dirname "${BASH_SOURCE[0]}" )" >/dev/null 2>&1 && pwd )"
 apt-get update && apt-get install -y git wget gcc g++
 
 apt-get install -y libglfw3 libgl1-mesa-glx libosmesa6 libglew-dev libsdl2-dev libsdl2-2.0-0
-apt-get install -y libegl-dev libegl
-apt-get install -y libglvnd0 libgl1 libglx0 libegl1 libgles2 xvfb
+apt-get install -y libglvnd0 libgl1 libglx0 libegl1 libgles2 xvfb libegl-dev libx11-dev freeglut3-dev
 
 git config --global --add safe.directory '*'
 root_dir="$(git rev-parse --show-toplevel)"
@@ -92,7 +91,7 @@ conda env config vars set \
 # make env variables apparent
 conda deactivate && conda activate "${env_dir}"
 
-pip install pip --upgrade
+# pip install pip --upgrade
 
 conda env update --file "${this_dir}/environment.yml" --prune
 #conda install -c conda-forge fltk -y

diff --git a/.github/unittest/linux_olddeps/scripts_gym_0_13/environment.yml b/.github/unittest/linux_olddeps/scripts_gym_0_13/environment.yml
@@ -2,7 +2,7 @@ channels:
   - pytorch
   - defaults
 dependencies:
-  - pip<25.0
+  - pip
   - protobuf
   - pip:
     - hypothesis

diff --git a/.github/unittest/linux_olddeps/scripts_gym_0_13/setup_env.sh b/.github/unittest/linux_olddeps/scripts_gym_0_13/setup_env.sh
@@ -15,11 +15,11 @@ printf "* Installing vim - git - wget\n"
 apt-get install -y vim git wget
 
 printf "* Installing glfw - glew - osmesa part 1\n"
-apt-get install -y libglvnd0 libgl1 libglx0 libegl1 libgles2 xvfb
+apt-get install -y libglvnd0 libgl1 libglx0 libegl1 libgles2 xvfb libx11-dev libegl-dev
+
+#printf "* Installing glfw - glew - osmesa part 2\n"
+#apt-get install -y libglfw3 libgl1-mesa-glx libosmesa6 libglew-dev libsdl2-dev libsdl2-2.0-0
 
-printf "* Installing glfw - glew - osmesa part 2\n"
-apt-get install -y libglfw3 libgl1-mesa-glx libosmesa6 libglew-dev libsdl2-dev libsdl2-2.0-0 libegl-dev libx11-dev
-#
 if [ "${CU_VERSION:-}" == cpu ] ; then
   # solves version `GLIBCXX_3.4.29' not found for tensorboard
 #    apt-get install -y gcc-4.9

diff --git a/.github/unittest/windows_optdepts/scripts/environment.yml b/.github/unittest/windows_optdepts/scripts/environment.yml
@@ -13,7 +13,6 @@ dependencies:
     - pytest-instafail
     - pytest-rerunfailures
     - expecttest
-    - pybind11[global]
     - pyyaml
     - scipy
     - coverage
diff --git a/.github/unittest/windows_optdepts/scripts/unittest.sh b/.github/unittest/windows_optdepts/scripts/unittest.sh
@@ -93,10 +93,11 @@ fi
 
 # install tensordict
 if [[ "$RELEASE" == 0 ]]; then
-  git clone https://github.com/pytorch/tensordict
-  cd tensordict
-  python setup.py develop
-  cd ..
+  conda install anaconda::cmake -y
+
+  python -m pip install "pybind11[global]"
+
+  python -m pip install git+https://github.com/pytorch/tensordict
 else
   pip3 install tensordict
 fi

diff --git a/.github/workflows/benchmarks.yml b/.github/workflows/benchmarks.yml
@@ -23,11 +23,19 @@ jobs:
 
   benchmark_cpu:
     name: CPU Pytest benchmark
-    runs-on: ubuntu-20.04
+    runs-on: linux.g5.4xlarge.nvidia.cpu
+    defaults:
+      run:
+        shell: bash -l {0}
+    container:
+      image: nvidia/cuda:12.3.0-base-ubuntu22.04
+      options: --cpus all
     steps:
       - name: Who triggered this?
         run: |
           echo "Action triggered by ${{ github.event.pull_request.html_url }}"
+      - name: Check ldd --version
+        run: ldd --version
       - name: Checkout
         uses: actions/checkout@v4
         with:
@@ -38,24 +46,54 @@ jobs:
           python-version: '3.10'
       - name: Setup Environment
         run: |
-          python3.10 -m venv ./py310
+          export TZ=Europe/London
+          export DEBIAN_FRONTEND=noninteractive  # tzdata bug
+          apt-get update -y
+          apt-get install software-properties-common cmake -y
+          add-apt-repository ppa:git-core/candidate -y
+          apt-get update -y
+          apt-get upgrade -y
+          apt-get -y install libglu1-mesa libgl1-mesa-glx libosmesa6 gcc curl g++ unzip wget libglfw3-dev libgles2-mesa-dev libglew-dev sudo git cmake libz-dev libpython3.10-dev
+      - name: Setup git
+        run: git config --global --add safe.directory /__w/rl/rl
+      - name: setup Path
+        run: |
+          echo /usr/local/bin >> $GITHUB_PATH
+      - name: Setup benchmarks
+        run: |
+          echo "BASE_SHA=$(echo ${{ github.event.pull_request.base.sha }} | cut -c1-8)" >> $GITHUB_ENV
+          echo "HEAD_SHA=$(echo ${{ github.event.pull_request.head.sha }} | cut -c1-8)" >> $GITHUB_ENV
+          echo "BASELINE_JSON=$(mktemp)" >> $GITHUB_ENV
+          echo "CONTENDER_JSON=$(mktemp)" >> $GITHUB_ENV
+          echo "PR_COMMENT=$(mktemp)" >>  $GITHUB_ENV
+      - name: Run
+        run: |
+          python3.10 -m venv --system-site-packages ./py310
           source ./py310/bin/activate
+          export PYTHON_INCLUDE_DIR=/usr/include/python3.10
 
-          python3 -m pip install --pre torch --index-url https://download.pytorch.org/whl/nightly/cpu -U 
-          python3 -m pip install git+https://github.com/pytorch/tensordict 
-          python3 setup.py develop
-          python3 -m pip install pytest pytest-benchmark
-          python3 -m pip install "gym[accept-rom-license,atari]"
-          python3 -m pip install "dm_control" "mujoco"
+          python3.10 -m pip install --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/cu128 -U
+          python3.10 -m pip install ninja pytest pytest-benchmark mujoco dm_control "gym[accept-rom-license,atari]"
+          python3 -m pip install "pybind11[global]"
+          python3.10 -m pip install git+https://github.com/pytorch/tensordict 
+          python3.10 setup.py develop
+
+          # test import
+          python3 -c """import torch
+          assert torch.cuda.device_count()
+          """
 
           cd benchmarks/
           export TORCHDYNAMO_INLINE_INBUILT_NN_MODULES=1
           export COMPOSITE_LP_AGGREGATE=0
+          export CUDA_VISIBLE_DEVICES=
           export TD_GET_DEFAULTS_TO_NONE=1
           python3 -m pytest -vvv --rank 0 --benchmark-json output.json --ignore test_collectors_benchmark.py
       - name: Store benchmark results
         uses: benchmark-action/github-action-benchmark@v1
         if: ${{ github.ref == 'refs/heads/main' || github.event_name == 'workflow_dispatch' }}
+        env:
+          GIT_WORK_TREE: /__w/rl/rl
         with:
           name: CPU Benchmark Results
           tool: 'pytest'
@@ -68,7 +106,6 @@ jobs:
           gh-pages-branch: gh-pages
           auto-push: true
 
-
   benchmark_gpu:
     name: GPU Pytest benchmark
     runs-on: linux.g5.4xlarge.nvidia.gpu
@@ -121,7 +158,8 @@ jobs:
           export PYTHON_INCLUDE_DIR=/usr/include/python3.10
 
           python3.10 -m pip install --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/cu128 -U
-          python3.10 -m pip install cmake ninja pytest pytest-benchmark mujoco dm_control "gym[accept-rom-license,atari]"
+          python3.10 -m pip install ninja pytest pytest-benchmark mujoco dm_control "gym[accept-rom-license,atari]"
+          python3 -m pip install "pybind11[global]"
           python3.10 -m pip install git+https://github.com/pytorch/tensordict 
           python3.10 setup.py develop
 

diff --git a/.github/workflows/benchmarks_pr.yml b/.github/workflows/benchmarks_pr.yml
@@ -14,11 +14,30 @@ jobs:
 
   benchmark_cpu:
     name: CPU Pytest benchmark
-    runs-on: ubuntu-20.04
+    runs-on: linux.g5.4xlarge.nvidia.cpu
+    defaults:
+      run:
+        shell: bash -l {0}
+    container:
+      image: nvidia/cuda:12.4.1-cudnn-runtime-ubuntu22.04
+      options: --cpus all
     steps:
+      - name: Set GITHUB_BRANCH environment variable
+        run: |
+          if [ "${{ github.event_name }}" == "push" ]; then
+            export GITHUB_BRANCH=${{ github.event.branch }}
+          elif [ "${{ github.event_name }}" == "pull_request" ]; then
+            export GITHUB_BRANCH=${{ github.event.pull_request.head.ref }}
+          else
+            echo "Unsupported event type"
+            exit 1
+          fi
+          echo "GITHUB_BRANCH=$GITHUB_BRANCH" >> $GITHUB_ENV
       - name: Who triggered this?
         run: |
           echo "Action triggered by ${{ github.event.pull_request.html_url }}"
+      - name: Check ldd --version
+        run: ldd --version
       - name: Checkout
         uses: actions/checkout@v4
         with:
@@ -27,28 +46,50 @@ jobs:
         uses: actions/setup-python@v4
         with:
           python-version: '3.10'
+      - name: Setup Environment
+        run: |
+          export TZ=Europe/London
+          export DEBIAN_FRONTEND=noninteractive  # tzdata bug
+          apt-get update -y
+          apt-get install software-properties-common cmake -y
+          add-apt-repository ppa:git-core/candidate -y
+          apt-get update -y
+          apt-get upgrade -y
+          apt-get -y install libglu1-mesa libgl1-mesa-glx libosmesa6 gcc curl g++ unzip wget libglfw3-dev libgles2-mesa-dev libglew-dev sudo git cmake libz-dev libpython3.10-dev
+      - name: Setup git
+        run: git config --global --add safe.directory /__w/rl/rl
+      - name: setup Path
+        run: |
+          echo /usr/local/bin >> $GITHUB_PATH
       - name: Setup benchmarks
         run: |
           echo "BASE_SHA=$(echo ${{ github.event.pull_request.base.sha }} | cut -c1-8)" >> $GITHUB_ENV
           echo "HEAD_SHA=$(echo ${{ github.event.pull_request.head.sha }} | cut -c1-8)" >> $GITHUB_ENV
           echo "BASELINE_JSON=$(mktemp)" >> $GITHUB_ENV
           echo "CONTENDER_JSON=$(mktemp)" >> $GITHUB_ENV
           echo "PR_COMMENT=$(mktemp)" >>  $GITHUB_ENV
-      - name: Setup Environment and tests
+      - name: Run
         run: |
-          python3.10 -m venv ./py310
+          python3.10 -m venv --system-site-packages ./py310
           source ./py310/bin/activate
+          export PYTHON_INCLUDE_DIR=/usr/include/python3.10
+
+          python3.10 -m pip install --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/cu128 -U
+          python3.10 -m pip install ninja pytest pytest-benchmark mujoco dm_control "gym[accept-rom-license,atari]"
+          python3.10 -m pip install "pybind11[global]"
+          python3.10 -m pip install git+https://github.com/pytorch/tensordict 
+          python3.10 setup.py develop
+          # python3.10 -m pip install git+https://github.com/pytorch/rl@$GITHUB_BRANCH
+
+          # test import
+          python3 -c """import torch
+          assert torch.cuda.device_count()
+          """
 
-          python3 -m pip install --pre torch --index-url https://download.pytorch.org/whl/nightly/cpu -U 
-          python3 -m pip install git+https://github.com/pytorch/tensordict 
-          python3 setup.py develop
-          python3 -m pip install pytest pytest-benchmark
-          python3 -m pip install "gym[accept-rom-license,atari]"
-          python3 -m pip install "dm_control" "mujoco"
-
           cd benchmarks/
           export TORCHDYNAMO_INLINE_INBUILT_NN_MODULES=1
           export COMPOSITE_LP_AGGREGATE=0
+          export CUDA_VISIBLE_DEVICES=
           export TD_GET_DEFAULTS_TO_NONE=1
           RUN_BENCHMARK="python3 -m pytest -vvv --rank 0 --ignore test_collectors_benchmark.py --benchmark-json "
           git checkout ${{ github.event.pull_request.base.sha }}
@@ -57,6 +98,8 @@ jobs:
           $RUN_BENCHMARK ${{ env.CONTENDER_JSON }}
       - name: Publish results
         uses: apbard/pytest-benchmark-commenter@v3
+        env:
+          GIT_WORK_TREE: /__w/rl/rl
         with:
           token: ${{ secrets.GITHUB_TOKEN }}
           benchmark-file: ${{ env.CONTENDER_JSON }}
@@ -130,7 +173,8 @@ jobs:
           export PYTHON_INCLUDE_DIR=/usr/include/python3.10
 
           python3.10 -m pip install --pre torch torchvision --index-url https://download.pytorch.org/whl/nightly/cu128 -U
-          python3.10 -m pip install cmake ninja pytest pytest-benchmark mujoco dm_control "gym[accept-rom-license,atari]"
+          python3.10 -m pip install ninja pytest pytest-benchmark mujoco dm_control "gym[accept-rom-license,atari]"
+          python3.10 -m pip install "pybind11[global]"
           python3.10 -m pip install git+https://github.com/pytorch/tensordict 
           python3.10 setup.py develop
           # python3.10 -m pip install git+https://github.com/pytorch/rl@$GITHUB_BRANCH