Amar Mudrankit
Amar Mudrankit
Reviewing again with @ahamino perspective: 1. This needs to be worker safe .. like works with custom_model_workers - I think it will be worker safe - if we fork the...
Just curious - how much RAM did you allocate for model when running with 5 workers.
Regarding improving MLOps library: 1. This is where I initially suggested - we need to run monitoring agent for the prediction gateway - It solves all the problems - lesser...
> > Regarding improving MLOps library: > > > > 1. This is where I initially suggested - we need to run monitoring agent for the prediction gateway - It...
> That would require us to redundantly parse requests and responses adding extra overhead. Not sure which parsing you are referring to? The library will write to a spool (say...
> > @yura-ant API spooler did not exist for 5 years. It was built within last 2 years. I think you meant async spooler. > > looks like from 2020...
Can we test how async reporting behaves in case of 429 or 5XX? If the async reporting could not recover from those errors, then we rather should not merge this...
> import requests ModuleNotFoundError: No module named 'requests' I suspect some issue with virtual environment - most likely - but yeah not sure. > if we make async api spooler...