

Improved hardware capabilities used to come very quickly (see Moore’s Law and Dennard Scaling). However that trend is basically over, so getting higher performance hardware takes a lot of effort to make hardware specialized for certain tasks. That’s why you see there inference accelerators like Groq, SambaNova, Cerebrus, etc. However this is hardware that still is gonna go into data centers. Something innovative has to happen on the AI side for commercial-grade models to be runnable on consumer hardware.










Just fyi, I tried one your instance. Searched a user, clicked a result, and got an error.
Error ./app.lua:134: attempt to concatenate field 'username' (a nil value) Traceback stack traceback: ./app.lua:134: in function 'handler' ...ittygram/lua_modules/share/lua/5.1/lapis/application.lua:185: in function 'resolve' ...ittygram/lua_modules/share/lua/5.1/lapis/application.lua:216: in function <...ittygram/lua_modules/share/lua/5.1/lapis/application.lua:214> [C]: in function 'xpcall' ...ittygram/lua_modules/share/lua/5.1/lapis/application.lua:214: in function 'dispatch' /apps/kittygram/lua_modules/share/lua/5.1/lapis/nginx.lua:231: in function 'serve' content_by_lua(nginx.conf.compiled:92):2: in main chunk