Qwen QwQ 32B on M4 Mac Mini Pro

Video Description

In this video, I put the 32-billion parameter QwQ-32B language model from Qwen through its paces on a base model Mac Mini M4 Pro with 24GB RAM and 512GB SSD. Useful Links: Install Ollama and DeepSeek R1 on Mac: https://www.youtube.com/watch?v=BzA0akucEg8&t Ollama: https://ollama.com/ Qwen on GitHub: https://qwenlm.github.io/blog/qwq-32b/ Gear I Use: https://www.head4space.com/gear/ Key points covered: • Walkthrough of QwQ using Ollama • Real performance testing with various prompts • Response time analysis with verbose mode • Practical demonstrations including RAG using MSTY The model shows surprisingly good performance, with response times ranging from 1-52 seconds depending on the task complexity. Even running alongside screen recording software, which uses some GPU resources, the model maintains usable performance levels. This video demonstrates that QwQ-32B is usable on a 24GB Mac Mini M4 Pro, despite using up the majority of the system resources. For those interested in running AI models locally, this is a viable option. Thanks to our YouTube community for suggesting this test! If you're interested in running AI locally, check out our other videos on language models.