Fix 'conda: not found' during Miniconda installation by updating PATH within the RUN command

Dockerfile

@@ -25,10 +25,12 @@ RUN apt-get update -y && apt-get install -qqy \
     && rm -rf /var/lib/apt/lists/* \
     && git lfs install # Initialize LFS system-wide
 
-# Install Miniconda (retain our existing approach)
+# Install Miniconda (REVISED: Add $CONDA_DIR/bin to PATH within this RUN command)
 RUN wget --quiet https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh -O miniconda.sh && \
     /bin/bash miniconda.sh -b -p $CONDA_DIR && \
     rm miniconda.sh && \
+    # Add Conda to PATH for subsequent commands in this RUN layer
+    export PATH=$CONDA_DIR/bin:$PATH && \
     conda clean --all --yes && \
     conda config --set auto_activate_base false && \
     conda config --add channels conda-forge
@@ -44,7 +46,7 @@ COPY . /app
 # Create the Conda environment named 'cosmos-predict1' using the provided YAML file.
 RUN conda env create -f cosmos-predict1.yaml
 
-# Set the default Conda environment to be activated and update PATH
+# Set the default Conda environment to be activated and update PATH (for subsequent layers and runtime)
 ENV CONDA_DEFAULT_ENV=cosmos-predict1
 ENV PATH=$CONDA_DIR/envs/cosmos-predict1/bin:$PATH
 
@@ -58,8 +60,6 @@ RUN . $CONDA_DIR/etc/profile.d/conda.sh && \
         --index-url https://download.pytorch.org/whl/cu121
 
 # IMPORTANT: Symlink fix for Transformer Engine compilation.
-# The `nvidia/cuda` base images place CUDA libraries and headers in /usr/local/cuda.
-# We need to ensure that the build system can find cuDNN headers.
 ENV CONDA_PREFIX_FIX=/usr/local/cuda
 RUN ln -sf $CONDA_PREFIX_FIX/lib/python3.10/site-packages/nvidia/*/include/* $CONDA_PREFIX_FIX/include/ || true && \
     ln -sf $CONDA_PREFIX_FIX/lib/python3.10/site-packages/nvidia/*/include/* $CONDA_PREFIX_FIX/include/python3.10 || true
@@ -83,5 +83,22 @@ RUN . $CONDA_DIR/etc/profile.d/conda.sh && \
 # Make the start.sh script executable.
 RUN chmod +x /app/start.sh
 
+# --- Verification Steps ---
+RUN echo "Verifying Python and Conda installations..."
+RUN python --version
+RUN conda env list
+RUN echo "Verifying PyTorch and CUDA availability..."
+RUN conda run -n cosmos-predict1 python <<EOF
+import torch
+print('PyTorch Version: ' + torch.__version__)
+print('CUDA Available: ' + str(torch.cuda.is_available()))
+if torch.cuda.is_available():
+    print('CUDA Device Name: ' + torch.cuda.get_device_name(0))
+else:
+    print('CUDA Device Name: N/A')
+EOF
+RUN [ $? -eq 0 ] || echo "PyTorch verification failed. Check dependencies in cosmos-predict1.yaml."
+# --- End Verification Steps ---
+
 # Set the default command to run when the container starts.
 CMD ["/app/start.sh"]