
ابو جواد
4.4K posts

ابو جواد
@ab0jwad
لقد جربت زحمة الأعمال، وكثرة الإرهاق فوجدت الفراغ أصعب منهما بكثير





Switched from gemma 4-31B to @Alibaba_Qwen's new Qwen3.6-35B-A3B on my Mac Studio M4 Max. Same 36 GB RAM, same prompt. Throughput went from 15.8 tok/s to 66.2 tok/s. For the blog automation that means a 1,800-word post now takes ~35 seconds instead of ~2.5 minutes. Architecture: 35B total params, 3B active via MoE. Native vision-language. 262K context. Apache 2.0. About 20 GB on disk. Tested reasoning with a plain word problem: "Train A leaves 9:00 AM at 80 km/h. Train B leaves 10:30 AM from 500 km away at 120 km/h toward A. When do they meet?" answer: 12:24 PM. step by-step deduction, 2,080 tokens, 30 seconds. Used jundot's oQ4 quant, loaded straight into oMLX, no conversion. six hermes @NousResearch profiles swapped, Gemma kept on dsk as a backup for nw.




















