Merge branch 'skypilot-org:master' into master

Dockerfile_k8s

@@ -1,4 +1,4 @@
-FROM continuumio/miniconda3:23.3.1-0
+FROM --platform=linux/amd64 continuumio/miniconda3:23.3.1-0
 
 # TODO(romilb): Investigate if this image can be consolidated with the skypilot
 #  client image (`Dockerfile`)
@@ -33,21 +33,15 @@ ENV HOME /home/sky
 # Set current working directory
 WORKDIR /home/sky
 
-# Install SkyPilot pip dependencies preemptively to speed up provisioning time
-RUN conda init && \
-    pip install wheel Click colorama cryptography jinja2 jsonschema networkx  \
-    oauth2client pandas pendulum PrettyTable rich tabulate filelock packaging  \
-    'protobuf<4.0.0' pulp pycryptodome==3.12.0 docker kubernetes==28.1.0  \
-    grpcio==1.51.3 python-dotenv==1.0.1 ray[default]==2.9.3 && \
+# Install skypilot dependencies
+RUN conda init && export PIP_DISABLE_PIP_VERSION_CHECK=1 && \
+    python3 -m venv ~/skypilot-runtime && \
+    PYTHON_EXEC=$(echo ~/skypilot-runtime)/bin/python && \
+    $PYTHON_EXEC -m pip install 'skypilot-nightly[remote,kubernetes]' 'ray[default]==2.9.3' 'pycryptodome==3.12.0' && \
+    $PYTHON_EXEC -m pip uninstall skypilot-nightly -y && \
     curl -LO "https://dl.k8s.io/release/v1.28.11/bin/linux/amd64/kubectl" && \
-    sudo install -o root -g root -m 0755 kubectl /usr/local/bin/kubectl
-
-# Add /home/sky/.local/bin/ to PATH
-RUN echo 'export PATH="$PATH:$HOME/.local/bin"' >> ~/.bashrc
-
-# Copy SkyPilot code base. This is required for the ssh jump pod to find the
-# lifecycle management scripts
-COPY --chown=sky . /skypilot/sky/
+    sudo install -o root -g root -m 0755 kubectl /usr/local/bin/kubectl && \
+    echo 'export PATH="$PATH:$HOME/.local/bin"' >> ~/.bashrc
 
 # Set PYTHONUNBUFFERED=1 to have Python print to stdout/stderr immediately
 ENV PYTHONUNBUFFERED=1

Dockerfile_k8s_gpu

@@ -41,19 +41,14 @@ RUN curl https://repo.anaconda.com/miniconda/Miniconda3-py310_23.11.0-2-Linux-x8
     eval "$(~/miniconda3/bin/conda shell.bash hook)" && conda init && conda config --set auto_activate_base true && conda activate base && \
     grep "# >>> conda initialize >>>" ~/.bashrc || { conda init && source ~/.bashrc; } && \
     rm Miniconda3-Linux-x86_64.sh && \
-    pip install wheel Click colorama cryptography jinja2 jsonschema networkx  \
-    oauth2client pandas pendulum PrettyTable rich tabulate filelock packaging  \
-    'protobuf<4.0.0' pulp pycryptodome==3.12.0 docker kubernetes==28.1.0  \
-    grpcio==1.51.3 python-dotenv==1.0.1 ray[default]==2.9.3 && \
+    export PIP_DISABLE_PIP_VERSION_CHECK=1 && \
+    python3 -m venv ~/skypilot-runtime && \
+    PYTHON_EXEC=$(echo ~/skypilot-runtime)/bin/python && \
+    $PYTHON_EXEC -m pip install 'skypilot-nightly[remote,kubernetes]' 'ray[default]==2.9.3' 'pycryptodome==3.12.0' && \
+    $PYTHON_EXEC -m pip uninstall skypilot-nightly -y && \
     curl -LO "https://dl.k8s.io/release/v1.28.11/bin/linux/amd64/kubectl" && \
-    sudo install -o root -g root -m 0755 kubectl /usr/local/bin/kubectl
-
-# Add /home/sky/.local/bin/ to PATH
-RUN echo 'export PATH="$PATH:$HOME/.local/bin"' >> ~/.bashrc
-
-# Copy SkyPilot code base. This is required for the ssh jump pod to find the
-# lifecycle management scripts
-COPY --chown=sky . /skypilot/sky/
+    sudo install -o root -g root -m 0755 kubectl /usr/local/bin/kubectl && \
+    echo 'export PATH="$PATH:$HOME/.local/bin"' >> ~/.bashrc
 
 # Set PYTHONUNBUFFERED=1 to have Python print to stdout/stderr immediately
 ENV PYTHONUNBUFFERED=1

docs/source/reference/faq.rst

@@ -192,6 +192,22 @@ For example, if you have access to special regions of GCP, add the data to ``~/.
 Also, you can update the catalog for a specific cloud by deleting the CSV file (e.g., ``rm ~/.sky/catalogs/<schema-version>/gcp.csv``).
 SkyPilot will automatically download the latest catalog in the next run.
 
+Package Installation
+---------------------
+
+Unable to import PyTorch in a SkyPilot task.
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+For `PyTorch <https://pytorch.org/>`_ installation, if you are using the default SkyPilot images (not passing in `--image-id`), ``pip install torch`` should work.
+
+But if you use your own image which has an older NVIDIA driver (535.161.08 or lower) and you install the default PyTorch, you may encounter the following error:
+
+.. code-block:: bash
+
+  ImportError: /home/azureuser/miniconda3/lib/python3.10/site-packages/torch/lib/../../nvidia/cusparse/lib/libcusparse.so.12: undefined symbol: __nvJitLinkComplete_12_4, version libnvJitLink.so.12
+
+You will need to install a PyTorch version that is compatible with your NVIDIA driver, e.g., ``pip install torch --index-url https://download.pytorch.org/whl/cu121``.
+
+
 Miscellaneous
 -------------