Skip to content

Releases: AparajeetS/Traintools

TrainTools 0.6.2

Choose a tag to compare

@AparajeetS AparajeetS released this 16 Jul 21:53

TrainTools 0.6.2: MCP Registry Namespace Correction

This metadata-only patch preserves the case-sensitive GitHub namespace required
by MCP Registry OIDC authentication. It includes the registry-compatible
traintools mcp command introduced in 0.6.1:

uvx --with mcp traintools mcp

TrainTools 0.6.1

Choose a tag to compare

@AparajeetS AparajeetS released this 16 Jul 21:45

TrainTools 0.6.1: Official MCP Registry Support

TrainTools can now be discovered by MCP-compatible agents and launched as a
local stdio server without a persistent installation:

uvx --with mcp traintools mcp

This patch release adds the official MCP Registry manifest, PyPI ownership
verification, and a stable traintools mcp command. The server remains local,
read-only, and focused on recommending and explaining training diagnostics.

TrainTools 0.6.0

Choose a tag to compare

@AparajeetS AparajeetS released this 16 Jul 18:11

TrainTools 0.6.0: Diagnostics That Humans And Agents Can Route Correctly

TrainTools now has a machine-readable diagnostic registry and a CLI that maps a
specific training problem to relevant tools:

pip install traintools
traintools recommend "validation loss plateaued" --json
traintools integration train-guard --framework huggingface

The release also adds structured JSON reports, focused problem guides,
AGENTS.md, llms.txt, and an optional local MCP server:

pip install "traintools[mcp]"
traintools-mcp

The recommender is transparent keyword routing, not an AI diagnosis. Each result
includes requirements, call timing, citations, and limitations, and it abstains
when no supported diagnostic matches.

traintools 0.5.1

Choose a tag to compare

@AparajeetS AparajeetS released this 28 Jun 00:10

Patch release for the PyPI README/install command.

  • Makes the primary install command pip install traintools.
  • Adds an explicit GitHub source link near the top of the README.
  • Keeps the paper-backed diagnostics released in 0.5.0.

Verification before release:

  • python -m pytest -q -> 84 passed
  • python -m build -> built wheel and sdist
  • python -m twine check dist\traintools-0.5.1* -> passed
  • Installed the built wheel in a temporary venv and imported top-level tools.