Training locally (using the command line)

Owned by Isaac Godfried
Last updated: Jun 12, 2023
1 min read

In this document we will go over how to use the command line to train instead of a notebook.

Running a single training

Create your configuration file (as shown in the examples). Then `python

Running a sweep

Running a sweep is a bit more complicated using the command line.

You will first need to create a configuration file base_config.json and a sweep_config.yaml. The entrypoint for the sweep should be run_flow.py or whatever you name it in step 2.

  1. You will then need to download the script located here and name it run_flow.py or whatever.

  2. Set BASE_CONFIG_PATH to the config path

  3. Create a sweep using the config you just created: wandb sweep --project sweeps_demo sweep_config.yaml copy the sweep_id

  4. wandb agent sweep_id (using the sweep_id you just created).