requirements.txt

gradio==4.26.0
transformers==4.38.1
torch==2.1.2
torchvision==0.16.2
torchaudio==2.1.2

# Optimization für HF Spaces
accelerate==0.27.0
bitsandbytes==0.42.0  # 8-bit quantization
peft==0.8.2
safetensors==0.4.1

# Model formats
sentencepiece==0.2.0
pydantic==2.5.3

# Server
python-multipart==0.0.9
fastapi==0.109.0  # Für app_fastapi.py
uvicorn==0.27.0   # ASGI server

# Optional: GGUF Support (für extrem komprimierte Modelle)
# llama-cpp-python==0.2.45  # Uncomment falls du GGUF nutzen willst

# Optional: Quantization Tools
# auto-gptq==0.7.1          # Für GPTQ quantization
# optimum==1.18.0           # Hugging Face Optimum