Commentary on X
@mervenoyann @thinkymachines @UnslothAI At 1-bit precision, you've basically bet everything on binary weights. 30-40 TPS for something that trades most of its accuracy for speed, the real speedrun is measuring how wrong it still is.
6:13 AM · Jul 16, 2026@mervenoyann @thinkymachines @UnslothAI That 1-bit quant speed is seriously impressive! And good shout on the llama.cpp webUI features; the reasoning slider sounds like a great addition for exploring model behavior.
7:36 AM · Jul 16, 2026@mervenoyann @thinkymachines @UnslothAI Getting that speed from a 1-bit quant locally is impressive.
8:04 AM · Jul 16, 2026@mervenoyann @thinkymachines @UnslothAI inkling being open weights from mira muratis new lab is huge for local dev work
7:01 AM · Jul 16, 2026View all
@mervenoyann @thinkymachines @UnslothAI Look at the Opensource AI ecosystem, we're becoming stronger by the day 🦾
5:51 AM · Jul 16, 2026Inkling @thinkymachines 1-bit quant by @UnslothAI running 30-40 TPS 🤯 video sped up, feel free to jump at the end to check outputs also llama.cpp webui is 😍 has reasoning slider, html preview, mcp and multimodal support
5:45 AM · Jul 16, 2026@mervenoyann @thinkymachines @UnslothAI I was also very surprised it was quite fast haha!
7:17 AM · Jul 16, 2026@danielhanchen @thinkymachines @UnslothAI it's a giant for sure 👨🏻🎤
7:31 AM · Jul 16, 2026