AI Trainer

A real, general-purpose VS Code extension for training and chatting with local AI models - pick a corpus file, configure the training/chat command for whatever you're actually running, start/stop a real process, and watch live output. It does not implement training itself; it's a real UI over commands you configure, so it works with LoRA fine-tuning, full fine-tuning, or anything else you can invoke from a shell.

Status: early, functional, actively developed. Built by Bachlaude, open source under the MIT License (see LICENSE).

Why

Training a small local model usually means juggling a terminal, a scattered set of corpus files, and remembering the right command-line flags each time. This extension puts that in one place: a real sidebar panel with model/corpus selection, one-click start/stop, and live output - without assuming you're using one specific framework or project layout.

Features

• Train tab: pick a model and corpus, choose a training method (LoRA / Full Fine-Tune / Custom), start/stop a real tracked process, watch live stdout/stderr.
• Corpus tab: create a new, blank corpus file to paste into directly from any AI chat (Claude, ChatGPT, Gemini, Copilot, Ollama - anywhere you generate training examples), or build entries one at a time with a simple instruction/input/output form.
• Chat tab: send a prompt to your configured model and see the real reply, useful for sanity-checking a checkpoint without leaving VS Code.
• Model detection: automatically finds models in a configured project directory AND your local Ollama installation (if running); add anything else manually via "Add Model...".
• Every real button has a visible label - keyboard users can still move fast, but nothing requires memorizing a shortcut to discover it.

Install

1. Clone this repository.
2. npm install
3. npm run compile
4. npx @vscode/vsce package to produce a .vsix, then code --install-extension <file>.vsix - or press F5 in VS Code to run it in an Extension Development Host for testing.

Setup

Open Settings (Ctrl+,) and search "AI Trainer". At minimum, set:

• aiTrainer.workingDirectory - where your training commands should run from (defaults to your open workspace folder if left blank).
• aiTrainer.modelsDir / aiTrainer.corpusDir - where your models and corpus files live.
• aiTrainer.trainCommand / aiTrainer.chatCommand - the real shell commands for your own project. Defaults point at a working LoRA example command - edit them for your own setup.

See FAQ.md for real hardware requirements, warnings, and known issues found during development.

Disclaimer

This software is provided "as is," without warranty of any kind - see LICENSE for the full text. Training runs it starts are real processes on your machine; review any command you configure before running it, especially if copied from somewhere you don't fully trust.

Feedback and bug reports

Genuinely wanted - please open a GitHub Issue. See CONTRIBUTING.md for the bug report format this project uses.