Open ryanrudes opened 1 month ago
Hi @ryanrudes thanks for reporting this. My apologies you ran into this issue. Do you mind giving us a bit more information about your environment? Specifically, the weave
version you are using, as well as your wandb
username. It is possible that this issue has already been resolved in master (here), but if you are logging large images or very large (30+MB) traces this notably does not account for that. Thanks!
Hi, my trace is most definitely larger than 30 MB. I was not aware that that would be considered large. I am noting that these runs are also appearing as ongoing in the portal despite having finished. Also, although all of the data I uploaded appears in the web portal, some of it is missing when I export and download the tables.
I am using weave 0.50.13 and wandb 0.17.6. My wandb username is ryanrudes-dtxplus.
@ryanrudes Both of those issues, large traces erroring and the export not containing the full data, are ticketed and under active development. They should be resolved in the near future; this thread will be updated when they are merged and deployed. Thanks for raising these.
Also, a new version of the weave SDK was released yesterday, so updating the weave package will include the changes I mentioned that might alleviate your upsert_batch
issue.
At the end of a fairly long run of
Evaluation.evaluate
(approx. 8000 examples), I get a peculiar error that does not appear for shorter runs.