Unverified50% confidenceBenchmarkExact time
使用SGLang同时开启Prefix Cache和MTP后，token生成速度提升到约40-50 tokens/s
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
8/18/2026
First Seen
Valid until: 11/16/2026
Sources
Related Claims
Unverified同等参数的Dense模型在没有MTP的情况下，生成速度通常只有40-50 tokens/s73% similarUnverifiedIn prefill speed tests, Halo achieved approximately 650 tokens/s while Spark achieved approximately 2000 tokens/s, with Spark being nearly three times faster.70% similarUnverifiedUnder a 4×RTX 3090 SXM4 configuration, the quantized version of GLM-5.2 achieves a token generation speed of approximately 6.3–6.8 tokens/second and a prompt evaluation speed of approximately 10.6 tokens/second.70% similarUnverifiedEach round of tool calls generates an average of 2K tokens of input and output, so by the 20th round, the tool call history alone accumulates approximately 40K tokens.66% similarUnverifiedIn the STM32 embedded development test, the locally deployed DS4 model had an average output speed of approximately 23 tokens/second, memory usage spiked to 110 GB during inference, and the peak speed reached 83.8 tokens/second.64% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/768406API
curl https://kongchang.com/api/v1/knowledge/claims/768406MCP
get_claim(id=768406)