用的newapi接的Nvidia 的deepseek v4pro，关闭了远程压缩，用的ccswitch 。无法自动压缩压缩上下文 
但是一旦上下文快满了就会报错 

stream disconnected before completion: Requested token count exceeds the model’s maximum context length of 262144 tokens. You requested a total of 262167 tokens: 196631 tokens from the input messages and 65536 tokens for the completion. Please reduce the number of tokens in the input messages or the completion to fi...