⚡️ Qwen3.8-Flash-Next just hit 325 tok/s on dual PRO 6000s. NVFP4 + MTP K4 on … · @QwenAIFans · TwKit

@QwenAIFans

⚡️ Qwen3.8-Flash-Next just hit 325 tok/s on dual PRO 6000s. NVFP4 + MTP K4 on stock vLLM: 127.8 → 325.4 tok/s on code-v1 (LocalMaxxing verified). Reasoning run landed at 258.4. Pro hardware, but the MTP jump is the part to steal. via LocalMaxxing 👇 https://www.localmaxxing.com/en/runs/cmvxyr5td07jioa01wy20ty4k

曝光 312 · 评论 0 · 点赞 4 · 书签 2