search

elixir-nx / bumblebee

Pre-trained Neural Network models in Axon (+ 🤗 Models integration)
Apache License 2.0
1.27k stars 90 forks source link

Compute BLIP image embeddings only once during generation #283

Closed jonatanklosko closed 7 months ago

jonatanklosko commented 7 months ago

Makes a difference when running with backend, but none when compiling with EXLA, so it's probably already falling into a graph optimisation.