# =============================================================================
# Book Recommendation System — Python Dependencies
# =============================================================================
#
# Recommended usage:
#   pip install -r requirements.txt
#
# This installs:
#   - Core backend (FastAPI + RAG + RecSys)
#   - Evaluation + dev tools (pytest, ruff, benchmark helpers)
#
# Optional heavy fine-tuning / LoRA / SFT dependencies are listed at the bottom
# in a separate commented section and can be installed only when needed.
# =============================================================================

# --- Base / production dependencies (API + RAG + RecSys) ---------------------

# API
fastapi>=0.109.0,<0.116.0
uvicorn[standard]>=0.27.0
pydantic>=2.0.0,<3.0.0
pydantic-settings>=2.0.0

# Data handling
pandas>=2.0.0
numpy>=1.24.0,<2.0.0
python-dotenv>=1.0.0

# LangChain / RAG stack
langchain>=0.2.0
langchain-community>=0.2.0
langgraph>=0.2.0
langchain-huggingface>=0.0.3
langchain-openai>=0.1.0

# ML / NLP
transformers>=4.40.0
torch>=2.0.0
sentence-transformers>=2.2.2
onnxruntime>=1.16.0
gensim>=4.3.0
lightgbm>=4.0.0
xgboost>=2.0.0
shap>=0.45.0
scikit-learn>=1.3.0
scipy>=1.11.0

# Infrastructure
redis>=5.0.0
huggingface-hub>=0.23.0
requests>=2.28.0
prometheus-client>=0.19.0
tqdm>=4.65.0

# Vector DB
faiss-cpu>=1.7.0

# OpenAI / LLM client
openai>=1.0.0


# --- Development / testing dependencies --------------------------------------

pytest>=7.0.0
pytest-cov>=4.0.0
ruff>=0.1.0
httpx>=0.25.0


# --- Optional extras: fine-tuning / LoRA / SFT / ColBERT --------------------
#
# These are only needed for:
#   - zero_shot / marketing fine-tuning scripts
#   - alternative reranker backends (e.g. ColBERT)
#
# Install manually when needed, e.g.:
#   pip install datasets accelerate peft trl bitsandbytes modelscope
#

# FastText backend for intent classifier (INTENT_BACKEND=fasttext)
# fasttext

# ColBERT reranker (RERANKER_BACKEND=colbert, lower latency)
# llama-index-postprocessor-colbert-rerank

# Fine-tuning (LoRA/SFT scripts in marketing/ and zero_shot/)
# datasets>=2.14.0
# accelerate>=0.26.0
# peft>=0.4.0
# trl>=0.7.0
# bitsandbytes>=0.41.0
# modelscope>=1.9.0