Mikhail Scherbina

I will try to train doppelgangers with superpoint + lightglue today

Got it also

I ended up with roll-my-own JWT token, works fine: ```go // main.go package main import ( "github.com/golang-jwt/jwt/v5" "github.com/joho/godotenv" "github.com/labstack/echo/v5" "github.com/pocketbase/pocketbase" "github.com/pocketbase/pocketbase/apis" "github.com/pocketbase/pocketbase/core" "github.com/pocketbase/pocketbase/models" "log" "os" "time" ) func main() {...

I am pro removing old code and just going along with the new version. Ifdefs can solve anything but those are just macros and do not provide static dispatch

There is an obvious problem for most people, using new versions of NPP, if it breaks older code, welp :(

Nice, would love to see it done!

Same

This means that code like ``` glm::vec3 n = glm::cross(pts[2] - pts[0], pts[1] - pts[0]); glm::vec3 look_dir = args.look_at - args.camera_pos; float intensity = dot(n, look_dir); if (intensity < 0)...

same, want to see quantized models running

Same thing. It would be nice to see something hierrarchical, or merging