Text Generation
PEFT
Safetensors
MLX
English
lora
mixture-of-experts
hobbylm
agent
macos
tool-use
computer-use
on-device
Instructions to use Hobbylm/hobbylm-1B-computer-use with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use Hobbylm/hobbylm-1B-computer-use with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("harims95/hobbylm-1B-instruct") model = PeftModel.from_pretrained(base_model, "Hobbylm/hobbylm-1B-computer-use") - MLX
How to use Hobbylm/hobbylm-1B-computer-use with MLX:
# Make sure mlx-lm is installed # pip install --upgrade mlx-lm # if on a CUDA device, also pip install mlx[cuda] # Generate text with mlx-lm from mlx_lm import load, generate model, tokenizer = load("Hobbylm/hobbylm-1B-computer-use") prompt = "Once upon a time in" text = generate(model, tokenizer, prompt=prompt, verbose=True) - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- MLX LM
How to use Hobbylm/hobbylm-1B-computer-use with MLX LM:
Generate or start a chat session
# Install MLX LM uv tool install mlx-lm # Generate some text mlx_lm.generate --model "Hobbylm/hobbylm-1B-computer-use" --prompt "Once upon a time"
- Atomic Chat
File size: 1,989 Bytes
6cf4e88 4ea1389 6cf4e88 4ea1389 6cf4e88 4ea1389 6cf4e88 1d3a733 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 | #!/usr/bin/env bash
# Hobby setup for Apple Silicon Macs. Downloads pinned model files, clones/builds the pinned macos-use server, builds Swift tools.
set -euo pipefail
cd "$(dirname "$0")/.."
mkdir -p data models vendor
python3 -m venv .venv
.venv/bin/pip install -q "mlx==0.32.3" "mlx-lm==0.31.3" "transformers==5.17.0" "tokenizers==0.23.2" "huggingface-hub==1.33.0" \
"numpy==2.5.3" "safetensors==0.8.0"
# FP32 Instruct weights (sha256 d0899e0a...); skipped if already present
[ -f models/hobbylm-1B-instruct/model.safetensors ] || .venv/bin/hf download harims95/hobbylm-1B-instruct \
--revision 8a90f6eb6a3d7c7eb952fd01b512a406821366ab --local-dir models/hobbylm-1B-instruct
# MLX runtime (executed code) + tokenizer, pinned revision; the 3.7 GB MLX weights are skipped (Hobby runs FP32 + adapter)
[ -d models/hobbylm-1B-mlx/runtime ] || .venv/bin/hf download harims95/hobbylm-1B-mlx \
--revision a2192174572dca075f8378db70cb3e8e70bfe521 --local-dir models/hobbylm-1B-mlx \
--include 'runtime/*' --include '*.json' --include '*.txt' # one --include per pattern; extra bare names would be read as files
[ -d models/hobbylm-1B-mlx/runtime/mlx_hobbylm ] || { echo "MLX runtime download failed (models/hobbylm-1B-mlx/runtime)" >&2; exit 1; }
# macos-use (Business Source License 1.1, see its LICENSE), pinned commit
[ -d vendor/mcp-server-macos-use ] || git clone https://github.com/mediar-ai/mcp-server-macos-use vendor/mcp-server-macos-use
git -C vendor/mcp-server-macos-use checkout -q 205c8880e8c06fa96a3e3a859e6edca728eb6c47
(cd vendor/mcp-server-macos-use && swift build -c release)
# small Swift tools: checkers and system controls
for t in calc_checker brightness media_key checker; do swiftc -O hobby/$t.swift -o data/$t; done
swiftc -O hobby/HobbyBar.swift -o data/HobbyBar # Hobby bar: Spotlight-style panel on Shift-Command-H (⇧⌘H)
echo "Done. Grant Accessibility to your terminal app, then: data/HobbyBar (⇧⌘H), or .venv/bin/python hobby/hobby_demo.py r4"
|