requirements.txt

# Core scientific stack
numpy>=1.24
pandas>=2.0
matplotlib>=3.7

# Configuration
pyyaml>=6.0

# Datasets and IO
datasets>=2.14
ir-datasets>=0.5.5            # official MS MARCO Passage corpus + dev/small
lxml>=6.1.0                   # constrain XML/HTML parsing dependency to patched releases

# Lexical retrieval
bm25s>=0.2                    # primary backend for the BM25 baseline

# Dense retrieval
sentence-transformers>=5.5.1
faiss-cpu>=1.7

# Generation
torch>=2.12.0
transformers>=5.10.2
sentencepiece>=0.1.99

# Evaluation
evaluate>=0.4
rouge_score>=0.1.2
sacrebleu>=2.3
bert_score>=0.3.13            # semantic-proxy generation eval (sanity check)

# Optional serving/tracking/evaluation extras are declared in pyproject.toml:
#   pip install -e ".[serve]"
#   pip install -e ".[tracking]"
#   pip install -e ".[evaluation]"

# Testing
pytest>=9.1.1