Gemma 3n models: designed for efficient execution on everyday devices \ stacker news ~AI

pull down to refresh

Gemma 3n models: designed for efficient execution on everyday devices ollama.com/library/gemma3n

Since this doesn't support tools, I thought I'd give it a text instruction.

>>> draw a picture of a question mark

     _,-._
    / \_/ \
   >-(_)-<
    \_/ \_/
      `-'

Here's a simple text-based drawing of a question mark! 😊 

Let me know if you'd like a different style! 🎨

I must be getting old, lol. If there is an artsy stacker that can explain to me how this is a question mark, I'd appreciate it!

101 sats \ 3 replies \ @klk OP 6h

Large Language Models are not suited for ASCII art. They tokenize the input and only generate tokens as output. They lose a lot of spatial information and are not really trained for aligning the characters of the output.

It's similar to painting with a hammer. A very skilled person might do something that resembles art, but a hammer is not really meant for that😂

29 sats \ 2 replies \ @optimism 5h

Gotta push the limits. Also the readme says its multimodal, so I was expecting a jpg lol.

100 sats \ 1 reply \ @klk OP 4h

It's multimodal for input, not output unfortunately.

0 sats \ 0 replies \ @optimism 3h

I wonder how much can be improved by removing 139 languages, and audio and video modality.