Skip to content
HearthworkHearthwork

Your first model

Download a model from Hugging Face with a repository name or a link:

Terminal window
hearthwork model unsloth/Qwen3-Coder-30B-A3B-Instruct-GGUF                    # a repo: lists files, you pick one
hearthwork model https://huggingface.co/<org>/<repo>                           # same, as a link
hearthwork model https://huggingface.co/<org>/<repo>/blob/main/<file>.gguf    # one file
  • The repo list shows each file’s size, smallest first, and whether it fits: in GPU memory (fastest), in GPU memory plus RAM (slower, fine for MoE models), or too big.
  • Files go to <models folder>/<org>/<repo>/, which is LM Studio’s layout. Multi-part models download all parts, and interrupted downloads resume when you run the command again.
  • For gated or private repositories, set the HF_TOKEN environment variable.

Context

Section titled “Context”

With context: auto, Hearthwork reads the model file (trained context, layers, KV heads, attention layout) and, together with your GPU memory and RAM, works out a range for each model. For example: 128K (auto; range 64K-256K, model max 256K).

  • The minimum is 64K, because agents’ own prompts are large.
  • The recommended size is what auto starts with, at most 128K.
  • The maximum is what fits in GPU memory plus RAM.

hearthwork context [model] [N|auto] shows the range or saves a size for a model (96K, 98304). The order of precedence is: start --context first, then the size saved for the model, then server.context in config.json, then auto.

Custom chat templates

Section titled “Custom chat templates”

If a model’s chat template is broken in some other way, put a fixed copy at templates/<model file name without .gguf>.jinja in the data folder.

Next: run your first agent session.