Configure Dockerfile with provided cosmos-predict1.yaml and install pip deps

Dockerfile

@@ -1,15 +1,70 @@
-FROM elungky/gen3c:latest
+# Start from a clean NVIDIA CUDA base image.
+# This provides the necessary CUDA runtime and development tools.
+# Using 12.4.0-devel-ubuntu22.04 to align with the CUDA version specified in your cosmos-predict1.yaml.
+FROM nvidia/cuda:12.4.0-devel-ubuntu22.04
 
+# Set environment variables for non-interactive installations to prevent prompts during apt-get.
+ENV DEBIAN_FRONTEND=noninteractive
+# Define the base directory for Conda installation.
+ENV CONDA_DIR=/opt/conda
+# Add Conda's binary directory to the system's PATH.
+ENV PATH=$CONDA_DIR/bin:$PATH
+
+# Set the working directory inside the container. All subsequent commands will run from here.
 WORKDIR /app
 
+# Install essential system dependencies required for Miniconda and general build processes.
+# This includes wget for downloading, git for cloning (if needed), build-essential for compiling,
+# and libgl1-mesa-glx for graphics-related libraries often used by ML frameworks.
+RUN apt-get update && apt-get install -y --no-install-recommends \
+    wget \
+    git \
+    build-essential \
+    libgl1-mesa-glx \
+    # Clean up apt cache to reduce image size
+    && rm -rf /var/lib/apt/lists/*
+
+# Install Miniconda:
+# 1. Download the Miniconda installer script.
+# 2. Run the installer in batch mode (-b) and specify the installation path (-p).
+# 3. Remove the installer script to save space.
+# 4. Clean Conda's package cache to further reduce image size.
+# 5. Configure Conda to not auto-activate the base environment on shell startup.
+# 6. Add the 'conda-forge' channel for broader package availability.
+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 && \
+    conda clean --all --yes && \
+    conda config --set auto_activate_base false && \
+    conda config --add channels conda-forge
+
+# Copy all local project files into the container's working directory (/app).
+# This includes your cosmos-predict1.yaml, gui/requirements.txt, start.sh, etc.
 COPY . /app
 
-# Install dependencies specific to the GUI/inference server
-# This assumes 'gui/requirements.txt' is directly under the '/app' directory after COPY.
-RUN pip install -r gui/requirements.txt
+# Create the Conda environment named 'cosmos-predict1' using the provided YAML file.
+# This step will install all specified Python, PyTorch, CUDA, and pip dependencies.
+RUN conda env create -f cosmos-predict1.yaml
+
+# Set the default Conda environment to be activated.
+ENV CONDA_DEFAULT_ENV=cosmos-predict1
+# Add the newly created Conda environment's binary directory to the PATH.
+# This ensures that executables (like python, pip, uvicorn) from this environment are found.
+ENV PATH=$CONDA_DIR/envs/cosmos-predict1/bin:$PATH
+
+# --- Verification Steps (Optional, but highly recommended for debugging) ---
+# These commands help confirm that Python, Conda, and PyTorch are set up correctly.
+RUN echo "Verifying Python and Conda installations..."
+RUN python --version
+RUN conda env list
+RUN echo "Verifying PyTorch and CUDA availability..."
+# Use 'conda run' to explicitly run the command within the 'cosmos-predict1' environment
+RUN conda run -n cosmos-predict1 python -c "import torch; print(f'PyTorch Version: {torch.__version__}'); print(f'CUDA Available: {torch.cuda.is_available()}'); print(f'CUDA Device Name: {torch.cuda.get_device_name(0) if torch.cuda.is_available() else 'N/A'}')" || echo "PyTorch verification failed. Check dependencies in cosmos-predict1.yaml."
+# --- End Verification Steps ---
 
-# Ensure start.sh is executable
+# Make the start.sh script executable.
 RUN chmod +x /app/start.sh
 
-# Use start.sh as the entrypoint for your Space
+# Set the default command to run when the container starts.
+# This will execute your start.sh script.
 CMD ["/app/start.sh"]

cosmos-predict1.yaml

@@ -1,22 +1,9 @@
-# SPDX-FileCopyrightText: Copyright (c) 2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
-# SPDX-License-Identifier: Apache-2.0
-#
-# Licensed under the Apache License, Version 2.0 (the "License");
-# you may not use this file except in compliance with the License.
-# You may obtain a copy of the License at
-#
-# http://www.apache.org/licenses/LICENSE-2.0
-#
-# Unless required by applicable law or agreed to in writing, software
-# distributed under the License is distributed on an "AS IS" BASIS,
-# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
-# See the License for the specific language governing permissions and
-# limitations under the License.
-
-# conda env create --file cosmos-predict1.yaml
 name: cosmos-predict1
 channels:
+  - pytorch          
+  - nvidia            
   - conda-forge
+  - defaults
 dependencies:
   - python=3.10
   - pip=25.0
@@ -27,3 +14,16 @@ dependencies:
   - cuda=12.4
   - cuda-nvcc=12.4
   - cuda-toolkit=12.4
+  # Add PyTorch and Torchvision
+  - pytorch=2.5.0a0 
+  - torchvision
+  - torchaudio
+  - pytorch-cuda=12.1 
+  # Add pip dependencies directly in the YAML
+  - pip:
+      - -r file:gui/requirements.txt 
+      - uvicorn[standard]
+      - fastapi
+      # Add any other crucial top-level pip dependencies here if your project needs them
+      # Example: - accelerate
+      # Example: - transformers

start.sh

@@ -1,61 +1,23 @@
 #!/bin/bash
-set -eux # Keep this for detailed logging
+set -eux # Keep this for debugging and seeing command execution
 
 export CUDA_VISIBLE_DEVICES="0"
 export CUDA_HOME="/usr/local/cuda"
 
-echo "Attempting to locate and activate Conda environment..."
-
-# Add a common Conda binary path to PATH (keeping this just in case, but it's failing)
-export PATH="/opt/conda/bin:$PATH"
-
-# --- DIAGNOSTIC STEPS: List contents of common Conda installation directories ---
-echo "Listing contents of /opt/:"
-ls -la /opt/ || echo "ls /opt/ failed or directory not found."
-echo "Listing contents of /usr/local/:"
-ls -la /usr/local/ || echo "ls /usr/local/ failed or directory not found."
-echo "Listing contents of /root/ (if accessible):"
-ls -la /root/ || echo "ls /root/ failed or directory not found (may be permission denied)."
-echo "--- END DIAGNOSTIC ---"
-
-# 1. Try to find the 'conda' executable in the system's PATH
-CONDA_EXEC=$(which conda)
-
-if [ -z "$CONDA_EXEC" ]; then
-    echo "ERROR: 'conda' executable still not found in PATH."
-    echo "This strongly suggests Conda is either not installed in the base image 'elungky/gen3c:latest', or it's in a highly unusual and non-standard location."
-    exit 1 # Exit here, no point in continuing if conda isn't found
-fi
-
-# The following lines will only execute if CONDA_EXEC is not empty
-echo "Found 'conda' executable at: $CONDA_EXEC"
-
-# 2. Derive the base Conda installation path from the executable's location.
-CONDA_BASE_PATH=$(dirname $(dirname "$CONDA_EXEC"))
-echo "Derived Conda base path: $CONDA_BASE_PATH"
-
-# 3. Construct the path to conda.sh script based on the derived base path
-CONDA_SH_PATH="$CONDA_BASE_PATH/etc/profile.d/conda.sh"
-
-if [ -f "$CONDA_SH_PATH" ]; then
-    echo "Found conda.sh at: $CONDA_SH_PATH"
-    source "$CONDA_SH_PATH" || { echo "ERROR: Failed to source conda.sh script at $CONDA_SH_PATH. Check permissions."; exit 1; }
-else
-    echo "ERROR: conda.sh not found at expected location derived from 'conda' executable: $CONDA_SH_PATH"
-    exit 1
-fi
-
-echo "Conda environment initialized successfully."
+echo "Activating Conda environment and starting server..."
 
 # Activate the specific conda environment
-conda activate cosmos-predict1 || { echo "ERROR: Failed to activate conda environment 'cosmos-predict1'. Ensure it exists and is accessible for this user."; exit 1; }
+# This now works because Conda is properly installed and in PATH by the Dockerfile
+source $CONDA_DIR/etc/profile.d/conda.sh # Source the conda.sh for the shell
+conda activate cosmos-predict1 # Activate the environment
 
-# Set PYTHONPATH after conda activation, as conda might adjust PATH/PYTHONPATH internally.
+# Set PYTHONPATH after conda activation for your specific project modules
 export PYTHONPATH="/app:/app/gui/api" 
 
 echo "Starting GEN3C FastAPI inference server..."
 
-export GEN3C_CKPT_PATH="/app/checkpoints"
+export GEN3C_CKPT_PATH="/app/checkpoints" # Assuming /app/checkpoints is still correct
 export GEN3C_GPU_COUNT=1
 
+# Start the FastAPI server
 exec uvicorn gui.api.server:app --host 0.0.0.0 --port 7860 --proxy-headers