jqhoogland / pattern-learning

Toy models of quanta learning

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Delayed Generalization: Unifying Grokking and Double Descent

Interpolating between grokking and double descent

Running the sweeps

Replace config.yml in the following with the relevant config file:

wandb sweep --project grokking <config.yml>

This will initialize a sweep.

To run the sweep, run the following command:

wandb agent <sweep_id>

where <sweep_id> is the id of the sweep you want to run. You can find the sweep id by running wandb sweep ls.

You can pass an optional --count flag to the wandb agent command to specify the number of runs you want to execute. If you don't pass this flag, the agent will run until all the runs in the sweep are complete (for a grid sweep).

On a multi-GPU machine, you can run multiple agents in parallel through the following:

CUDA_VISIBLE_DEVICES=0 wandb agent <sweep_id> &
CUDA_VISIBLE_DEVICES=1 wandb agent <sweep_id> &
...

Toy Model

See the jupyter notebooks in toy_models for more instructions.

About

Toy models of quanta learning


Languages

Language:Jupyter Notebook 77.1%Language:Python 22.9%