[KongchangAI]
Unverified50% confidenceFactExact time

mmBERT支持最高8192 tokens的输入窗口

1
Sources
50%
Confidence
Long-term
Relevance
9/12/2026
First Seen

Sources

mmBERT在CPU上加速推理的4个关键优化技巧

Related Entities

mmBERTmodel

Related Claims

Unverified在CPU上进行文本分类时，256或512 tokens的分块大小通常优于8192的大窗口67% similarUnverified一项日志研究测量出各Harness在读到用户提示词前发送的Token：OpenCode约7000 tokens，CloudCode约33000 tokens，Pi约1300 tokens65% similarUnverified许多免费或低价大语言模型的上下文窗口可能只有4K-8K tokens64% similarUnverified不同大模型在上下文窗口大小方面存在显著差异，从8K到128K tokens不等64% similar

Cite This Claim

Stable URI
https://kongchang.com/claim/912399
API
curl https://kongchang.com/api/v1/knowledge/claims/912399
MCP
get_claim(id=912399)