thomas.yaml 1.3 KB

1234567891011121314151617181920212223242526272829
  1. name: thomas
  2. repo: https://github.com/keppy/hermes-plugin-thomas
  3. sha: 7302ab947f99e3adb16e78e139794fe1696134f8
  4. description: >-
  5. Fine-tune a small calibrated encoder classifier on Modal and rate it with gonogo. Every GPU launch
  6. goes through the human approval gate with the exact config shown; data checks and evaluation run
  7. locally for free. Disclosure — thomas_encoder_train spends the user's Modal GPU credits and uploads
  8. the chosen training JSONL rows to Modal after the human approval gate; the base model is downloaded
  9. from Hugging Face on the Modal side. Training and inference run in a separate Python interpreter the
  10. user configures via THOMAS_PYTHON (with the author's thomas package, torch and modal installed); the
  11. Hermes venv only gets gonogo-eval. Runs and artifacts are written under HERMES_HOME/thomas/runs; the
  12. eval tool writes report files to the path the model supplies.
  13. maintainer: keppy
  14. tier: community
  15. category: tools
  16. requires_hermes: ">=0.21"
  17. platforms: [linux, macos]
  18. version: "0.1.3"
  19. docs_url: https://github.com/keppy/hermes-plugin-thomas#readme
  20. capabilities:
  21. provides_tools:
  22. - thomas_check_data
  23. - thomas_encoder_train
  24. - thomas_run_status
  25. - thomas_encoder_eval
  26. provides_hooks:
  27. - pre_tool_call
  28. provides_middleware: []
  29. requires_env: []