The error env.gpu-type

user: the script can change it: the error says what happened in the experiment’s names, and what to write instead.

Where MIME gives it, with what its message says happened and what its action says to do ({…} is filled in from the experiment):

• in mime/env/_gpus.py:

  • what happened: {…} asks for several GPUs; a MIME run computes on one or {…} is not a GPU type MIME knows as exact or the GPU type {…} may be served by another model of GPU, so its results name no one machine

  • what to do: request one, --gpu <type> without a count: {…} or request one of them with --gpu <type>: {…} or request it exactly: --gpu {…}

• in mime/env/_modal_script.py:

  • what happened: --deterministic sets XLA's deterministic mode for a GPU, and the run requests none

  • what to do: request a GPU, 'mime modal <folder> --gpu <type> --deterministic', or drop --deterministic for the CPU run

Reading an error and reaching its full traceback: reading an error. Every code: the index.