stanford-crfm / mistral

Mistral: A strong, northwesterly wind: Framework for transparent and accessible large-scale language model training, built with Hugging Face 🤗 Transformers.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Add Codalab Tutorial doc

dlwh opened this issue · comments

@teetone and @percyliang have done all the work, we should just make a doc of it and check it in.

Other related tasks:

  • Change Codalab to expose an env variable for job id, make mistral look for that env variable if one is not set.
  • minimize the number of flags needed to start a run (specifically deepspeed num_gpus, num_nodes flags)
  • extend tutorial to show how to reuse preprocessing cache