feat: Add Python packaging support for pip installation

[yurekami]

Summary

• Adds pyproject.toml for modern Python packaging (PEP 517/518)
• Adds module docstring and __version__ for better API
• Expands __all__ to export additional utility functions

Motivation

Currently, users must manually copy eplb.py to use this library. With proper packaging support, users can:

# Install from GitHub
pip install git+https://github.com/deepseek-ai/EPLB.git

# Or install for local development
git clone https://github.com/deepseek-ai/EPLB.git
cd EPLB
pip install -e .

Then use it in their code:

import eplb
print(eplb.__version__)  # "1.0.0"

phy2log, log2phy, logcnt = eplb.rebalance_experts(weight, num_replicas, num_groups, num_nodes, num_gpus)

Changes

New file: pyproject.toml

• Modern build system configuration (setuptools backend)
• Package metadata (name, version, description, authors)
• Python version requirement (>=3.8)
• PyTorch dependency (>=1.10.0)
• PyPI classifiers for discoverability

Modified: eplb.py

• Added module-level docstring
• Added __version__ = "1.0.0"
• Expanded __all__ to include:
  • __version__
  • rebalance_experts (main entry point)
  • rebalance_experts_hierarchical (hierarchical policy)
  • balanced_packing (utility for bin packing)
  • replicate_experts (utility for expert replication)

Test plan

• Verify pyproject.toml is valid
• Verify Python syntax is correct
• Test pip install -e . works locally
• Test import and basic usage

🤖 Generated with Claude Code