Replying to @dhh
I really like the idea, android and iOS app is almost ready here (with support of resizeable activity)
Demo in 4g (really far away from home, via tailscale)
1
1
70
Replying to @laurent_zw
Is this really better? With llama.cpp Ingram and Q4 unsloth I have :
ngram-map-k 7 (retenu) │ Vulkan0 │ 301 │ 54,0
(Halo Strix 128gb)
1
78
Replying to @liquidai
Nice , it makes my AI max+ Hermes compatible without too much latency.
But in my setup, the reasoning level seems always on regardless of the configuration.
Did I miss something?
1
3
437
Replying to @nicodotdev @liquidai
Nice , it makes my AI max+ Hermes compatible without too much latency.
But in my setup, the reasoning level seems always on regardless of the configuration.
Did I miss something?
84
Replying to @outsource_
Interesting, in my setup, I have the speed of the 27b (~12t/s Strix Halo 128gb).
The draft model seems loaded (visible in the logs)
1
1
264
En mode sauvage (j'ai tourné plusieurs années comme ça, commande volet de velux solaire).
L'avantage de la Tahoma mini c'est le retour d'état dans home assistant via homekit.
1
133












