ViewTube

I tested Nimble 9B and Tev1 4B, Ollama's new decision models (Jev-type) with a real dilemma: which Mac Studio to buy, the M5 ...

IA local en un iMac M1 de 8 GB: pruebo cinco modelos a 4 bits con oMLX para ver cuál rinde de verdad y dónde está el límite de ...

Local OpenClaw on a mini PC: I tested the Beelink SEi13 AI with a 24 GB NPU and Qwen 3.6 35B-A3B, which exceeded 27 tokens per ...

Qwen 3.5 running local AI on the MacBook Neo: I tested the 2B, 4B, and 9B models with MLX, and the only one that could write a ...

Qwen 3.8 27B on two 16GB GPUs: I went from 13 to 66 tokens per second. But two 16GB cards are not the same as one 32GB card ...

Qwen 3.8 27B en un Mac mini de 16 GB: cabe a 2 bits, va a 7,36 tokens por segundo, y a mitad de respuesta se rompe. Al final te ...

I tested Qwen 3.8 27B on a 48GB MacBook Pro M3 Max with different quantizations, and the differences are enormous. In LM ...

The RTX 5080 has more than double the memory bandwidth of a MacBook Pro M3 Max, and yet the Mac wins in local AI. I tested ...

Presentación de Hardware IA. ¿Qué hardware necesitas para mover IA en local? Aquí lo mido y te doy los números. Pruebo ...