Configure Space to run GEN3C FastAPI inference server

Dockerfile

@@ -1,17 +1,15 @@
-# Dockerfile
 FROM elungky/gen3c:latest
 
-# Set the working directory inside the container.
-# This is where your application will expect its files (e.g., /app/cosmos_predict1)
 WORKDIR /app
 
-# Copy all files from the current local directory (where your Dockerfile is)
-# into the /app directory inside the container.
-# This will ensure cosmos_predict1/, assets/, checkpoints/, and start.sh are all present.
 COPY . /app
 
-# Make the start.sh script executable inside the container
-#RUN chmod +x /app/start.sh
+# 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
 
-# Execute the startup script when the container runs
-#CMD ["/app/start.sh"]
+# Ensure start.sh is executable
+RUN chmod +x /app/start.sh
+
+# Use start.sh as the entrypoint for your Space
+CMD ["/app/start.sh"]

start.sh

@@ -1,45 +1,28 @@
 #!/bin/bash
-set -eux # This makes the script exit immediately on any command failure and prints commands as they execute. Very useful for debugging!
+set -eux # Keep this for debugging and seeing command execution
 
-# Set environment variables for a single GPU on Hugging Face Spaces
 export CUDA_VISIBLE_DEVICES="0"
-export CUDA_HOME="/usr/local/cuda"
-export PYTHONPATH="/app"
+export CUDA_HOME="/usr/local/cuda" # Robust CUDA_HOME for Docker environment
+export PYTHONPATH="/app"           # Set Python path to the app root
 
-echo "Starting GEN3C application on A100 Large 80GB GPU..."
+echo "Starting GEN3C FastAPI inference server..."
 
-# Create the /data/output directory for persistent storage
-mkdir -p /data/output
+# Set environment variables as per GEN3C guide
+# Assuming 'checkpoints' folder is directly under '/app'
+export GEN3C_CKPT_PATH="/app/checkpoints"
+export GEN3C_GPU_COUNT=1 # Explicitly use 1 GPU for A100 instance
 
-# Define a temporary output path for the video within the /app directory (container's working dir)
-TEMP_VIDEO_PATH="/app/temp_test_single_image.mp4"
+# Navigate to the 'gui' directory where server.py resides, if necessary
+# The current directory is /app, and server.py is at /app/gui/api/server.py
+# So we need to ensure the path is correct from /app
+# Or, you can CD into the directory: cd gui
+# For now, let's keep the path explicit.
 
-# Run your Python script. It will now try to save the video to the temporary path.
-python cosmos_predict1/diffusion/inference/gen3c_single_image.py \
-  --checkpoint_dir checkpoints \
-  --input_image_path assets/diffusion/000000.png \
-  --video_save_name "$TEMP_VIDEO_PATH" \ 
-  --guidance 1 \
-  --foreground_masking \
-  --offload_diffusion_transformer \
-  --offload_tokenizer \
-  --offload_text_encoder_model \
-  --offload_prompt_upsampler \
-  --offload_guardrail_models \
-  --disable_guardrail \
-  --disable_prompt_encoder
-
-# --- Post-execution checks ---
-# Check if the temporary video file was actually created by the Python script
-if [ -f "$TEMP_VIDEO_PATH" ]; then
-    echo "SUCCESS: Temporary video file was created at: $TEMP_VIDEO_PATH"
-    # If created, move it to the persistent storage location
-    mv "$TEMP_VIDEO_PATH" /data/output/test_single_image.mp4
-    echo "SUCCESS: Video moved to /data/output/test_single_image.mp4"
-else
-    echo "ERROR: Temporary video file NOT found at $TEMP_VIDEO_PATH after Python script execution."
-    echo "This indicates the Python script likely failed to generate the video."
-    exit 1 # Exit with an error code to make the Space red if the video isn't produced
-fi
-
-echo "Script finished successfully. Generated video is in /data/output/. Container is exiting."
+# Start the FastAPI server
+# Using 'uvicorn' directly is often preferred over 'fastapi dev' for deployment,
+# as 'fastapi dev' is more for development with auto-reload.
+# 'uvicorn' is typically installed with 'fastapi[standard]' or 'uvicorn[standard]'
+# The format is 'uvicorn <module>:<app_instance>'
+# So if server.py has 'app = FastAPI()', it's 'gui.api.server:app'
+# We need to bind to 0.0.0.0 and port 7860 for HF Spaces
+exec uvicorn gui.api.server:app --host 0.0.0.0 --port 7860 --proxy-headers