Jonathan K

[Jonathan K @jmklex]

@bisheshab holy shit

[Jonathan K @jmklex]

@teortaxesTex @leo_linsky My money is on MiMo I'd like to believe that Xiaomi cooked something up

[Jonathan K @jmklex]

@teortaxesTex @leo_linsky It's definitely Chinese (dumb questions I dislike the meme of sending Chinese models this but point stands)

[Jonathan K @jmklex]

@DJLougen How many active params?

[Jonathan K @jmklex]

How many active params?

[Daniel Lougen @DJLougen]

Next qwen model will be an 80b moe, you heard it here first

[Jonathan K @jmklex]

they just changed fable it talks normally now

[Jonathan K @jmklex]

@BitGrateful Microsoft Presidio is a great SDK for this. Local LLMs will get there, and be good for this use case quite soon, but right now the amount of compute you need to do this with a good model makes it inaccessible to most. Presidio allows you to do the same thing (also offline).