Usage
This guide is for day-to-day usage with one script:
SmallLanguageModel Usage (PowerShell)
This guide is for day-to-day usage with one script:
A) `index` (docs-folder mode)
Uses leumas-docs scanFolders (focuses on docs/doc folders).
B) `index-md` (filesystem mode, markdown or markdown+json)
Scans files under a root path (default extensions are .md/.markdown/.mdx).
Useful index props (noob version)
--docs-root What folder to scan. Example: D:\Leumas
4) Build Chunks (Recommended For Better RAG)
This builds chunks.jsonl, which powers stronger retrieval and citations.
5) Train a New Model (Fresh)
Use this when you want a new base model.
Important props
--model Name of the model bundle folder (saved in artifacts\models\<name).
6) Train a New Version from Existing Model
Use this when you want to keep old model and create a new one.
9) Output Files You Should Know
Index metadata: data/index/docsindex.json
Command says source model is missing tokenizer/checkpoint
That model is not fully trained yet. Run train-new first (with --overwrite if needed), then run train-next.