3 ms·
I've been writing my own, out of curiosity, with gemma4. I've been surprised how far I'm getting.
by jvalencia 5mo ago
I've been writing my own, out of curiosity, with gemma4. I've been surprised how far I'm getting.
- zambelli 5mo agoVery cool! Hopefully you'll share it someday!
- tmzt 4mo agoSame. I switched my efforts to a larger Gemma 4 MoE model (26B-A4B) and llama.cpp and started getting meaningful results. I also implemented subagents for querying, determining which object/action to execute, and composing a short title. This is all running on an M4 in approximately 16 gb of ram. Also using Google's native tool calling channels.