Demo repository for Inspect AI presentation in Vegas 2025
uv sync --locked
Then activate the virtual environment. Otherwise, you'll need to prepend uv run to the commands below:
source .venv/bin/activate
Set up your model provider auth (e.g., OpenAI):
export OPENAI_API_KEY=your-openai-api-key
For other providers, see https://inspect.aisi.org.uk/providers.html
Run MMLU eval:
inspect eval single_turn/mmlu.py --model openai/gpt-4
Run MMLU eval with eval-set command:
inspect eval-set single_turn/mmlu.py --model openai/gpt-4 \
--max-connections 20 \
--token-limit 2000 \
--epochs 5
We use the python eval call, so simply run the script.
python single_turn/bomb.py
Check out ./doomla/README.md
inspect view --log-dir assets/
jupyter notebook
Then connect to http://localhost:8888 and open the .ipynb files