[Q] Why does Wandb revert to a single dot at the end of training?
cinjon opened this issue · comments
When my axolotl model finishes finetuning, it becomes a single dot on wandb. I'm not doing anything special. My config looks like this, where the nulls are detritus from auto-saving configs:
wandb_entity: <entityname>
wandb_log_model: null
wandb_name: <unique_wandbname>
wandb_project: <projectname>
wandb_run_id: null
wandb_watch: null
In the wandb log file, it shows all the step-logs as expected.
But at the end it looks like this:
Hi @cinjon , what version of wandb are you currently using?
Are you at any point trying to write to your graph at a previous step, that already has data written to it?
Hi there, I wanted to follow up on this request. Please let us know if we can be of further assistance or if your issue has been resolved.
It hasn't been resolved; I've just been busy elsewhere. I'm not sure what's happening right now as it only happens with one type of training and not another. I apologize, but I'll come back to this soon.
Sounds good, waiting for an update.
Hi! Wanted to follow up with you regarding this thread
Hi, since we have not heard back from you we are going to close this request. If you would like to re-open the conversation, please let us know!