{"id":"2028749477519982739","url":"https://x.com/_orcaman/status/2028749477519982739","text":"Mac users: if you are running Qwen models locally, make sure you use MLX (Apple's version of PyTorch, sort of). It's the only way to get decent performance.\n\nWindows users: just buy a GeForce RTX ffs","author":{"name":"Or Hiltch","username":"_orcaman","avatarUrl":"https://pbs.twimg.com/profile_images/1992581933520318464/NJnsoD8__200x200.jpg"},"createdAt":"Tue Mar 03 08:29:02 +0000 2026","engagement":{"replies":18,"retweets":40,"likes":534,"views":66965},"quoteTweet":{"id":"2028694110782263365","url":"https://x.com/stevibe/status/2028694110782263365","text":"Update: Qwen3.5:9b head-to-head (MLX, Apple Silicon optimized)\n\nMac Studio M2 Ultra: 89.74 tok/s\nMac Mini M4: 20.82 tok/s\n\nMLX basically doubles the speed on both machines.","author":{"name":"stevibe","username":"stevibe","avatarUrl":"https://pbs.twimg.com/profile_images/1476819230557614081/dIp-8a5r_200x200.jpg"},"createdAt":"Tue Mar 03 04:49:01 +0000 2026"}}