requirements.txt

# Core UI
gradio>=4.36.0

# Custom frontend server
fastapi>=0.100.0
uvicorn>=0.23.0
python-multipart>=0.0.6
spaces>=0.50.0

# PDF extraction
PyMuPDF>=1.24.0

# Image generation (Flux2KleinPipeline requires >=0.37.0)
diffusers>=0.37.0

# LLM inference
torch>=2.3.0
transformers==4.57.1
accelerate>=0.31.0
sentencepiece>=0.2.0
protobuf>=4.25.0
bitsandbytes>=0.43.0

# llama.cpp runtime (PAPERPROF_RUNTIME=llamacpp) — prebuilt CPU wheel, no compile
--extra-index-url https://abetlen.github.io/llama-cpp-python/whl/cpu
llama-cpp-python>=0.3.0