jetstream-pytorch
jetstream-pytorch copied to clipboard
[Feature Request] Per request sampling params
Currently sampling params such as temperature are set as commandline flags in when the server starts.
It would be nice for each request to pass in the sampling params instead.