henrycunh

> Experimental support for ES Modules (bytecode/sourceless support not implemented yet) YES BRO, YES.

having the same problem here!

i'm currently developing a module for it :] it'll be a service layer generated from the OAS files, which you can import as a plugin in your `nuxt/auth` module configuration!

any way around this?

same here!

just use bun either [directly](https://bun.sh/docs/bundler/executables) or through [`bkg`](https://github.com/theseyan/bkg)! worked super well for me :]

perhaps something like this to preserve float precision on cuda? ```diff diff --git a/point_e/util/precision_compatibility.py b/point_e/util/precision_compatibility.py new file mode 100644 --- /dev/null +++ b/point_e/util/precision_compatibility.py @@ -0,0 +1,5 @@ +import torch +import...

> I apologize for my question, but how noticeable is the change to float32? I'm pretty confident that using higher precision, like `float64`, will almost always give us tighter, more...