• Implemented robust model and scene reference validations in both the FastAPI backend and frontend to prevent empty, corrupt, or missing files from silently failing or producing duplicated background
fallbacks.
• Enhanced scenery prompt translation on the backend to dynamically map user-specified "Image" numbers e.g. Image 1, Image 2 to Qwen-specific "Picture" placeholders Picture 1, Picture 2, avoiding template
mismatch desyncs.
Changes
• File & Path Validation: Added detailed validations to the /generate-scenery endpoint to ensure model_filename and scene_image paths exist on disk, are resolved without query parameters stripping ?t=..., are
not directories, and are openable and verifiable as PIL images.
• Identical Reference Checks: Introduced checks in both the backend and frontend submitGenerateScenery inside car.html to raise an HTTP 400 exception / display a toast warning if the model and scene references
point to the same image.
• Dynamic Prompt Mapping: Configured the backend to auto-translate occurrences of image 1/2/3 in custom scenery prompts to Picture 1/2/3, facilitating correct Qwen layout hint targeting.
• Regression Unit Tests: Wrote comprehensive unit tests test_13b_scenery_validation_checks in test_regression_api.py verifying file checking, duplicate reference detection, and query parameter stripping.
Verification
• Executed test_regression_api.py simulated API testing with 100% success 19/19 tests passed.
• Executed test_regression_qwen.py end-to-end model/hardware integration test suite successfully 2/2 tests passed.
• Implemented global offline mode for all HuggingFace-dependent modules to eliminate unauthenticated request warnings and prevent external connection attempts during startup and runtime.
Changes
• Global Offline Configuration: Added HF_HUB_OFFLINE=1 and TRANSFORMERS_OFFLINE=1 to the environment variables at the top of edit_api.py, embeddings.py, watcher.py, and pose_llm/pose_llm_api.py.
• Explicit Local Files Only: Updated all huggingface_hub.hf_hub_download calls in edit_api.py to include local_files_only=True, ensuring they never attempt to reach the Hub if models are already cached.
• LLM Service Optimization: Updated AutoTokenizer and AutoModelForCausalLM calls in pose_llm/pose_llm_api.py to use local_files_only=True.
• Consistency: Ensured that shared modules like embeddings.py which uses open_clip are also locked to offline mode to prevent background downloads.
• Implemented global offline mode for all HuggingFace-dependent modules to eliminate unauthenticated request warnings and prevent external connection attempts during startup and runtime.
Changes
• Global Offline Configuration: Added HF_HUB_OFFLINE=1 and TRANSFORMERS_OFFLINE=1 to the environment variables at the top of edit_api.py, embeddings.py, watcher.py, and pose_llm/pose_llm_api.py.
• Explicit Local Files Only: Updated all huggingface_hub.hf_hub_download calls in edit_api.py to include local_files_only=True, ensuring they never attempt to reach the Hub if models are already cached.
• LLM Service Optimization: Updated AutoTokenizer and AutoModelForCausalLM calls in pose_llm/pose_llm_api.py to use local_files_only=True.
• Consistency: Ensured that shared modules like embeddings.py which uses open_clip are also locked to offline mode to prevent background downloads.
