懒猫微服官方Single-stream decoding speed improved by 2%.
懒猫微服官方Single-stream decoding speed improved by 4.9%.
懒猫微服官方DeepSeek 1017 pre-filling performance improved by 8.3%.
懒猫微服官方DeepSeek V4.1 Flash 1017: decoding performance improved by 1.7%–5.2%, and 32K prefill performance increased by 3.6%.