# Grok-1's Mixture of Experts: Active Parameters Are Not Stored Parameters: runnable files

5 NumPy CPU tests of an eight-expert linear routing toy; no Grok checkpoint, inference benchmark or GPU run.

Download all files in this folder before running. The article is at https://oneruby.dev/unveiling-grok-1-the-frontier-of-ai-modeling/.

## Reproduce

Tested with Python 3.11.5 on arm64 macOS. Use an isolated environment; requirements.txt pins direct dependencies and requirements.lock records the tested dependency closure for this platform.

```sh
python3 -m venv .venv
.venv/bin/python -m pip install -r requirements.lock
.venv/bin/python -B -m unittest -v test_example.py
.venv/bin/python -B example.py
```

## Files

- [requirements.lock](requirements.lock)

- [example.py](example.py)
- [requirements.txt](requirements.txt)
- [test_example.py](test_example.py)

## Boundaries

- Linear NumPy toy, not the actual Grok architecture, checkpoint or serving system.
- All experts remain local; no sharding, capacity, offload, attention, KV cache or latency/memory benchmark.
- Selected-score softmax is explicitly the toy convention; not an implementation claim about every Grok routing detail.