# Core
torch>=2.1.0
transformers>=4.40.0
accelerate>=0.28.0
datasets>=2.18.0

# Quantization
auto-gptq>=0.7.0
autoawq>=0.2.0
bitsandbytes>=0.43.0

# Fine-tuning
peft>=0.10.0
trl>=0.8.0

# Inference
vllm>=0.4.0

# Evaluation
lighteval>=0.5.0

# LLM Judge (pick one)
openai>=1.12.0
anthropic>=0.25.0

# Analysis & Visualization
numpy>=1.24.0
pandas>=2.0.0
matplotlib>=3.7.0
seaborn>=0.13.0
scipy>=1.10.0

# Utilities
tqdm>=4.65.0
pyyaml>=6.0
jsonlines>=4.0.0
# difflib