fix: HfFolder ImportError - add compatibility shim for huggingface_hub 1.x

Fix: Always use FastAPI as server, remove Gradio-only fallback, remove demo.launch()

Mount Gradio at /gradio to prevent interference with API routes

Fix Request parameter to use dependency injection

Add file validation and fix request parameter order

Fix CORS for GitHub Pages and improve multipart request handling

Add detailed logging for multipart requests and route registration

Fix exception handler to prevent UnicodeDecodeError and improve error messages

Add route registration logging before Gradio mount

Add explicit JSONResponse to segment endpoint

Fix middleware to not consume request body for file uploads

Add request logging middleware to debug Method Not Allowed errors

Add route debugging and ensure FastAPI routes are properly registered

Fix API routes and add root endpoint

Add detailed logging for API requests to show in Hugging Face logs

Increase upload limit to 2 GB to match processing limit

Increase file size limits: upload 1 GB, processing 2 GB

Fix critical issues: 50MB upload limit, report guardrails, T2 experimental status, API security, and indentation errors

Fix UnboundLocalError and Triton cache permission error

Fix Internal Server Error: mount Gradio Blocks to FastAPI app for Hugging Face Spaces

Fix Gradio 4.44.x schema bug: patch additionalProperties boolean to dict conversion

Fix syntax error: remove embedded error messages from app.py

Fix Gradio launch error: remove unsupported api_name parameter from click() method

Fix No API found error: remove API info patching and explicitly set api_name

Fix API info generation to prevent No API found error

Fix remaining indentation error at line 133

Fix indentation error in log_startup_health function

Fix Gradio API info generation bug and localhost launch error

Fix TF32 API

Switch Dockerfile to CUDA 12.1 + cu121 (recommended), use venv, add nvcc check, verify extensions at build time

Add Python Space detection and nvcc check - warn user to switch to Docker Space for CUDA extensions

Docker: Add OpenCV libs, use PYTORCH_ALLOC_CONF only with max_split_size_mb=128, create postBuild for Python Spaces fallback

Fix: Remove unsupported max_split_size_mb=128, use PYTORCH_ALLOC_CONF instead of deprecated PYTORCH_CUDA_ALLOC_CONF

Fix: Exact import test format, TF32 set directly (no try/except), verify allocator config includes max_split_size_mb=128

Fix: PyTorch 2.9.0 cu128, simplified setup.sh with fatal errors, TF32=tf32, cuDNN always on, TORCH_CUDA_ARCH_LIST=8.9

Fixed

Production build strategy: Docker support, pinned dependencies, OOM guardrails, optimized defaults

Fix OOM: reduce ROI depth to 64, sw_batch_size=1, CPU aggregation, fix env var timing, update TF32 API

Major performance and stability improvements: disable torch.compile by default, GPU preprocessing, ROI auto-adjustment, timeout mechanism, validation, cuDNN benchmarking, version logging, concurrency limits

Optimize for dedicated GPU: Add L4 GPU support with higher memory thresholds and better batch sizes

Fix naming conflict: rename model.py to model_loader.py and initialize DEVICE to prevent NoneType errors

Refactor app.py into modular structure: config, model, processing, inference, and app modules

Add aggressive memory management: unload unused models and check memory before inference

Remove CPU fallback - model requires CUDA and cannot run on CPU

Add PyTorch memory allocator config and enhanced OOM recovery with CPU fallback

Remove all code comments and add adaptive memory management for OOM prevention

Add comprehensive diagnostic logging and performance optimizations

Revert: Restore overlap to 0.15 for better accuracy (matches original deployment)

Optimize: Reduce sliding window overlap and add inference progress tracking

Fix: Handle image_tensor cleanup in finally block safely