requirements.txt

# Core dependencies
torch>=2.2.0
transformers>=4.35.0
accelerate>=0.24.0
peft>=0.5.0
trl>=0.7.2
sentencepiece>=0.1.99
protobuf>=4.24.0
optimum>=1.14.0
scipy>=1.11.0
numpy>=1.24.0
pandas>=2.0.0
tqdm>=4.66.0
matplotlib>=3.7.0
seaborn>=0.12.0

# Environment management
python-dotenv>=1.0.0

# Unsloth for GRPO
unsloth>=0.4.0
flash-attn>=2.4.0

# Dataset handling
datasets>=2.14.0
jsonlines>=4.0.0
pyarrow>=14.0.0

# Tokenization
tiktoken>=0.5.0

# Evaluation and metrics
scikit-learn>=1.3.0
evaluate>=0.4.0

# API access
openrouter>=0.3.0
requests>=2.31.0
aiohttp>=3.8.0

# Experiment tracking
wandb>=0.15.0

# Documentation
pyyaml>=6.0.0
jinja2>=3.1.0

# Development tools
black>=23.9.0
isort>=5.12.0
pytest>=7.3.0

# MLX support (Apple Silicon optimization)
mlx>=0.0.1; platform_system == "Darwin"

# Optionals
jupyter>=1.0.0
ipywidgets>=8.1.0
fastapi>=0.103.0; python_version >= "3.8"

# For distributed training
ray>=2.6.0

# Kaggle API for dataset downloads
kaggle>=1.5.16