Training Commands

The buddy train group (buddy.cli.simple_train_cli) wraps the buddy.train API for one-line local fine-tuning. It is available when the training dependencies are installed:

pip install "buddy-ai[training]"

Commands

Command	Description
buddy train <data> --name <name>	Train a model on a file or directory
buddy train test <name>	Generate a sample from a trained model
buddy train list	List your trained models
buddy train delete <name>	Delete a trained model
buddy train models	List available open-source base models
buddy train install-deps	Install the training dependencies

Train

buddy train /path/to/data --name my-model
buddy train /path/to/data --name coder --model phi-2 --epochs 5

Option	Alias	Description
--name	-n	Required. Name for the trained model
--model	-m	Base model alias or full Hugging Face id
--epochs	-e	Training epochs (default 3)
--description	-d	Description stored in metadata
--force	-f	Overwrite an existing model

The data path is a positional argument and must exist. Models are saved under ~/.buddy/trained_models/<name> (see Model Training).

Test

buddy train test my-model --prompt "Hello, how are you?"
buddy train test my-model -l 150

Option	Alias	Description
--prompt	-p	Prompt to send (a default is used if omitted)
--max-length	-l	Number of tokens to generate (default 100)

List and delete

buddy train list
buddy train delete old-model        # prompts for confirmation
buddy train delete old-model --yes  # skip the prompt

delete accepts -y/--yes to skip confirmation.

Available base models

buddy train models

Prints the curated list of open-source base models and their aliases (e.g. distilgpt2, phi-2, mistral-7b, llama3-8b). Gated models such as Llama, Mistral, and Gemma require a Hugging Face login first.

Install dependencies

buddy train install-deps

Installs transformers, torch, datasets, and accelerate via pip — an alternative to the [training] extra.

See also

• Training Overview
• Data Preparation · Model Training