How-ToDevelopers
Jun 20, 8:23 AM
User shares tip to fit Qwen 3.6 27B at 131k context on 7900XTX
Tip uses CatchyOS with iGPU bypass and q5_0 kv cache to maximize VRAM. Achieves Q6K+MTP at 131k context on 24GB card.
·
Jun 20, 8:23 AM
Tip uses CatchyOS with iGPU bypass and q5_0 kv cache to maximize VRAM. Achieves Q6K+MTP at 131k context on 24GB card.