wf Module

Fortran interface to Weights & Biases (wandb) experiment tracking.

This is the user-facing API module. It re-exports the public symbols from the feature modules in src/wf/:

  • wf_run — wandb_init, wandb_log, wandb_finish, wandb_shutdown
  • wf_config — wandb_config_set, wandb_config_get
  • wf_sweep — wandb_sweep_config_type, wandb_sweep, wandb_agent, wandb_sweep_start_agent, wandb_sweep_next_params, wandb_sweep_run_done

Quick start

use wf

call wandb_init(project="my_project", name="run-01")
call wandb_config_set("learning_rate", 0.001d0)
call wandb_config_set("epochs", 100)

do epoch = 1, 100
   ! ... training ...
   call wandb_log("loss", loss_val, step=epoch)
end do

call wandb_finish()

Prerequisites

  • Python >= 3.8 with wandb installed (pip install wandb)
  • Link against libwandb_fortran (which itself links Python)

Uses