Load a model
Start with a public GGUF URL—no Ollama installation or local model download required.
byhand model --url <direct-gguf-url>
Model interpretability, made visual
See inside a language model.
Wondering what happens inside an LLM? Follow the architecture and math from your terminal.
Current system requirement: macOS on an Apple Silicon Mac (M1 or newer). Intel Macs, Windows, and Linux are not supported yet.
Bring your model: Works with Ollama, local GGUF files, and public GGUF URLs.
- Model sources Local files · Public URLs · Ollama
- Explains Architecture · Shapes · Operations
- Reads only Metadata and tensor descriptors—never weight values
How it works
From model file to understandable math.
Follow the computation
Move through attention, normalization, residual paths, tensor shapes, equations, and parameter sources.
Choose your depth
Survey the full architecture, focus on one operation, or trace how calculated values flow through the pipeline.
Explore and export
Use the view that fits the question.
Navigate the whole model
Step through operations and open a focused matrix explainer without leaving your terminal.
Share an interactive view
Export every step and explainer as one self-contained file that opens in a browser.
Trace the numerical pipeline
Follow formula-driven proxy tensors whose calculated outputs flow into later operations.
Get started
Install, then open a working model.
The fastest first run uses a public model URL, so you do not need Ollama or a downloaded GGUF file. Read the quick start
curl -fsSL https://raw.githubusercontent.com/imruljubair/byhand/main/install.sh | sh
byhand model --url https://huggingface.co/tensorblock/Llama-3.2-1B-Instruct-GGUF/resolve/main/Llama-3.2-1B-Instruct-Q2_K.gguf