5 10 hours ago

Qwen-3.5 9b (Dense not MoE) distilled with Deepseek 4 reasoning traces and customized params, average output rate (high/xhigh thinking).

ollama run ducquoc/qd-9b-1103

Models

View all →

Readme

Qwen 3.5 9b fine tuned using Deepseek 4 reasoning traces.

Capabilities: completion, tools, thinking. More suitable to run on GPU, rather than CPU (Mac M3 output ~ 25 token/s).

https://ducquoc.wordpress.com/2026/10/17/top-5-free-providers-deepseek-flash/