Ceyda Cinarel (재이다)

I can't give any production ready guarantees 😅 but a basic setup probably do something easily building on top of official containers https://hub.docker.com/r/pytorch/torchserve like this; *pseudo* dockerfile: ``` FROM pytorch/torchserve:latest-gpu...

I vaguely remember running into this problem elsewhere, an update of streamlit to the newest version should do the trick. I have just released an updated version with latest streamlit...

I'm a bit hesitant about this change; 1. I would rather not have `import streamlit` statement (ie front end) in api.py. In case anyone wants to use just the APIs....

Lets shelve this until there is a more concrete InferenceAPI. I opened this https://github.com/cceyda/torchserve-dashboard/pull/11 for some ideas

> I don't think we should implement a logic depending on the input type inside of the InferenceApi. > It should work no matter if the `input_` is a buffer,...

@shijianjian yes the formula is correct. Adding the option to pass a background color would be pretty handy, since it is often used. Making me wonder if we should also...

> EG is for parity with Kernel Gateway environments in which each user spawns their own instance of the gateway. I would rather avoid each user having to spin their...

aha, The EG I'm running is not contained right now (running on bare metal). so I guess it makes more sense to use localprocessproxy & dockerprocessproxy (custom launcher). whereas a...

> 1. EG can't spawn "contained" kernels w/o also being "contained" itself. Are you finding otherwise? I'm using a custom ProcessProxy for launching docker containers (custom images), which also adds...

> Nice. Are you thinking about contributing that back to the project at all? I keep meaning to do that but my timeline is chaotic. I thought better to start...