Update

[ghstack-poisoned]

Author: vmoens

.github/unittest/linux_libs/scripts_isaaclab/isaac.sh

@@ -20,6 +20,8 @@ export TF_CPP_MIN_LOG_LEVEL=0
 export BATCHED_PIPE_TIMEOUT=60
 export TD_GET_DEFAULTS_TO_NONE=1
 export OMNI_KIT_ACCEPT_EULA=yes
+export PIP_DISABLE_PIP_VERSION_CHECK=1
+export PYTHONNOUSERSITE=1
 
 nvidia-smi
 
@@ -46,36 +48,49 @@ eval "$(${conda_dir}/bin/conda shell.bash hook)"
 conda create --prefix ${env_dir} python=3.10 -y
 conda activate ${env_dir}
 
+# Set LD_LIBRARY_PATH to prioritize conda environment libraries
+export LD_LIBRARY_PATH=${lib_dir}:${LD_LIBRARY_PATH:-}
+
+# Install a compatible version of expat (< 2.6.0 to avoid XML_SetReparseDeferralEnabled symbol issues)
+conda install -c conda-forge "expat<2.6" -y
+
+# Reinstall Python to ensure it's properly linked against the conda expat
+conda install --force-reinstall python=3.10 -y
+
+# Verify the expat linkage
+echo "* Checking pyexpat linkage:"
+python -c "import pyexpat; print('pyexpat imported successfully')" || echo "WARNING: pyexpat import failed"
+
 # Pin pytorch to 2.5.1 for IsaacLab
 conda install pytorch==2.5.1 torchvision==0.20.1 pytorch-cuda=12.4 -c pytorch -c nvidia -y
 
-conda run -p ${env_dir} pip install --upgrade pip
-conda run -p ${env_dir} pip install 'isaacsim[all,extscache]==4.5.0' --extra-index-url https://pypi.nvidia.com
+python -m pip install --upgrade pip --disable-pip-version-check
+python -m pip install 'isaacsim[all,extscache]==4.5.0' --extra-index-url https://pypi.nvidia.com --disable-pip-version-check
 conda install conda-forge::"cmake>3.22" -y
 
 git clone https://github.com/isaac-sim/IsaacLab.git
 cd IsaacLab
-conda run -p ${env_dir} ./isaaclab.sh --install sb3
+./isaaclab.sh --install sb3
 cd ../
 
 # install tensordict
 if [[ "$RELEASE" == 0 ]]; then
   conda install "anaconda::cmake>=3.22" -y
-  conda run -p ${env_dir} python -m pip install "pybind11[global]"
-  conda run -p ${env_dir} python -m pip install git+https://github.com/pytorch/tensordict.git
+  python -m pip install "pybind11[global]" --disable-pip-version-check
+  python -m pip install git+https://github.com/pytorch/tensordict.git --disable-pip-version-check
 else
-  conda run -p ${env_dir} python -m pip install tensordict
+  python -m pip install tensordict --disable-pip-version-check
 fi
 
 # smoke test
-conda run -p ${env_dir} python -c "import tensordict"
+python -c "import tensordict"
 
 printf "* Installing torchrl\n"
-conda run -p ${env_dir} python -m pip install -e . --no-build-isolation
-conda run -p ${env_dir} python -c "import torchrl"
+python -m pip install -e . --no-build-isolation --disable-pip-version-check
+python -c "import torchrl"
 
 # Install pytest
-conda run -p ${env_dir} python -m pip install pytest pytest-cov pytest-mock pytest-instafail pytest-rerunfailures pytest-error-for-skips pytest-asyncio
+python -m pip install pytest pytest-cov pytest-mock pytest-instafail pytest-rerunfailures pytest-error-for-skips pytest-asyncio --disable-pip-version-check
 
 # Run tests
-conda run -p ${env_dir} python -m pytest test/test_libs.py -k isaac -s
+python -m pytest test/test_libs.py -k isaac -s

.github/unittest/linux_libs/scripts_minari/environment.yml

@@ -25,6 +25,6 @@ dependencies:
     - gymnasium-robotics
     - minari[create]
     - jax>=0.7.0
-    - mujoco<3.3.6      
+    - mujoco>=2.3.0,<3.0.0
     - mujoco-py<2.2,>=2.1
     - minigrid

.github/workflows/test-linux-libs.yml

@@ -91,40 +91,40 @@ jobs:
 
         bash .github/unittest/linux_libs/scripts_brax/run_all.sh
 
-  unittests-d4rl:
-    strategy:
-      matrix:
-        python_version: ["3.10"]
-        cuda_arch_version: ["12.8"]
-    if: ${{ github.event_name == 'push' || contains(github.event.pull_request.labels.*.name, 'Data') }}
-    uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
-    with:
-      repository: pytorch/rl
-      runner: "linux.g5.4xlarge.nvidia.gpu"
-      docker-image: "nvidia/cuda:12.4.0-devel-ubuntu22.04"
-      timeout: 120
-      script: |
-        if [[ "${{ github.ref }}" =~ release/* ]]; then
-          export RELEASE=1
-          export TORCH_VERSION=stable
-        else
-          export RELEASE=0
-          export TORCH_VERSION=nightly
-        fi
+  # unittests-d4rl:
+  #   strategy:
+  #     matrix:
+  #       python_version: ["3.10"]
+  #       cuda_arch_version: ["12.8"]
+  #   if: ${{ github.event_name == 'push' || contains(github.event.pull_request.labels.*.name, 'Data') }}
+  #   uses: pytorch/test-infra/.github/workflows/linux_job_v2.yml@main
+  #   with:
+  #     repository: pytorch/rl
+  #     runner: "linux.g5.4xlarge.nvidia.gpu"
+  #     docker-image: "nvidia/cuda:12.4.0-devel-ubuntu22.04"
+  #     timeout: 120
+  #     script: |
+  #       if [[ "${{ github.ref }}" =~ release/* ]]; then
+  #         export RELEASE=1
+  #         export TORCH_VERSION=stable
+  #       else
+  #         export RELEASE=0
+  #         export TORCH_VERSION=nightly
+  #       fi
 
-        set -euo pipefail
-        export PYTHON_VERSION="3.10"
-        export CU_VERSION="cu128"
-        export TAR_OPTIONS="--no-same-owner"
-        export UPLOAD_CHANNEL="nightly"
-        export TF_CPP_MIN_LOG_LEVEL=0
-        export BATCHED_PIPE_TIMEOUT=60
-        export TD_GET_DEFAULTS_TO_NONE=1
+  #       set -euo pipefail
+  #       export PYTHON_VERSION="3.10"
+  #       export CU_VERSION="cu128"
+  #       export TAR_OPTIONS="--no-same-owner"
+  #       export UPLOAD_CHANNEL="nightly"
+  #       export TF_CPP_MIN_LOG_LEVEL=0
+  #       export BATCHED_PIPE_TIMEOUT=60
+  #       export TD_GET_DEFAULTS_TO_NONE=1
 
-        bash .github/unittest/linux_libs/scripts_d4rl/setup_env.sh
-        bash .github/unittest/linux_libs/scripts_d4rl/install.sh
-        bash .github/unittest/linux_libs/scripts_d4rl/run_test.sh
-        bash .github/unittest/linux_libs/scripts_d4rl/post_process.sh
+  #       bash .github/unittest/linux_libs/scripts_d4rl/setup_env.sh
+  #       bash .github/unittest/linux_libs/scripts_d4rl/install.sh
+  #       bash .github/unittest/linux_libs/scripts_d4rl/run_test.sh
+  #       bash .github/unittest/linux_libs/scripts_d4rl/post_process.sh
 
   unittests-envpool:
     strategy:
@@ -156,10 +156,10 @@ jobs:
         export BATCHED_PIPE_TIMEOUT=60
         export TD_GET_DEFAULTS_TO_NONE=1
 
-        bash .github/unittest/linux_libs/scripts_d4rl/setup_env.sh
-        bash .github/unittest/linux_libs/scripts_d4rl/install.sh
-        bash .github/unittest/linux_libs/scripts_d4rl/run_test.sh
-        bash .github/unittest/linux_libs/scripts_d4rl/post_process.sh
+        bash .github/unittest/linux_libs/scripts_envpool/setup_env.sh
+        bash .github/unittest/linux_libs/scripts_envpool/install.sh
+        bash .github/unittest/linux_libs/scripts_envpool/run_test.sh
+        bash .github/unittest/linux_libs/scripts_envpool/post_process.sh
 
   unittests-gendgrl:
     strategy:

docs/source/reference/envs.rst

@@ -1130,6 +1130,7 @@ to be able to create this other composition:
     InitTracker
     KLRewardTransform
     LineariseRewards
+    ModuleTransform
     MultiAction
     NoopResetEnv
     ObservationNorm

examples/distributed/collectors/multi_nodes/delayed_dist.py

@@ -150,7 +150,7 @@ def make_env():
         if i == 10:
             t0 = time.time()
     t1 = time.time()
-    torchrl_logger.info(f"time elapsed: {t1-t0}s, rate: {counter/(t1-t0)} fps")
+    torchrl_logger.info(f"time elapsed: {t1 - t0}s, rate: {counter / (t1 - t0)} fps")
     collector.shutdown()
     exit()
 

examples/distributed/collectors/multi_nodes/delayed_rpc.py

@@ -148,7 +148,7 @@ def make_env():
         if i == 10:
             t0 = time.time()
     t1 = time.time()
-    torchrl_logger.info(f"time elapsed: {t1-t0}s, rate: {counter/(t1-t0)} fps")
+    torchrl_logger.info(f"time elapsed: {t1 - t0}s, rate: {counter / (t1 - t0)} fps")
     collector.shutdown()
     exit()
 

examples/distributed/collectors/multi_nodes/generic.py

@@ -125,5 +125,5 @@ def gym_make():
             t0 = time.time()
     collector.shutdown()
     t1 = time.time()
-    torchrl_logger.info(f"time elapsed: {t1-t0}s, rate: {counter/(t1-t0)} fps")
+    torchrl_logger.info(f"time elapsed: {t1 - t0}s, rate: {counter / (t1 - t0)} fps")
     exit()

examples/distributed/collectors/multi_nodes/rpc.py

@@ -113,5 +113,5 @@ def gym_make():
             t0 = time.time()
     collector.shutdown()
     t1 = time.time()
-    torchrl_logger.info(f"time elapsed: {t1-t0}s, rate: {counter/(t1-t0)} fps")
+    torchrl_logger.info(f"time elapsed: {t1 - t0}s, rate: {counter / (t1 - t0)} fps")
     exit()

examples/distributed/collectors/multi_nodes/sync.py

@@ -119,5 +119,5 @@ def gym_make():
             t0 = time.time()
     collector.shutdown()
     t1 = time.time()
-    torchrl_logger.info(f"time elapsed: {t1-t0}s, rate: {counter/(t1-t0)} fps")
+    torchrl_logger.info(f"time elapsed: {t1 - t0}s, rate: {counter / (t1 - t0)} fps")
     exit()

examples/distributed/collectors/single_machine/generic.py

@@ -160,5 +160,5 @@ def gym_make():
             t0 = time.time()
     collector.shutdown()
     t1 = time.time()
-    torchrl_logger.info(f"time elapsed: {t1-t0}s, rate: {counter/(t1-t0)} fps")
+    torchrl_logger.info(f"time elapsed: {t1 - t0}s, rate: {counter / (t1 - t0)} fps")
     exit()