| """AutoSLM — managed LoRA post-training: log in with your freesolo key, train. | |
| A focused developer experience (TOML run specs, pluggable environments, | |
| CLI/API/MCP entry points, adapter deployment). Users authenticate with their | |
| freesolo API key (`slm login`); the control plane runs each job on a managed | |
| GPU (RunPod or Vast.ai) behind the scenes. | |
| """ | |
| __all__ = ["__version__"] | |
| __version__ = "0.2.0" | |
| # Back-compat: HF_TOKEN is the canonical name, but the README/.env.example promise the legacy | |
| # HUGGINGFACE_TOKEN is still accepted. The code was renamed to HF_TOKEN with no shim, so an operator | |
| # who set only HUGGINGFACE_TOKEN would get NO token forwarded (silent auth failure). Normalize here, | |
| # at package import (covers worker, CLI, serve, and the control plane), to honor the documented alias. | |
| import os as _os | |
| if _os.environ.get("HUGGINGFACE_TOKEN") and not _os.environ.get("HF_TOKEN"): | |
| _os.environ["HF_TOKEN"] = _os.environ["HUGGINGFACE_TOKEN"] | |