DeepSeek Using context after 28-30 makes each request take a very long time and is extremely slow.

[ai-llt]

Question

At first, I thought it was an issue with the API service provider, until I suddenly noticed that one client was still waiting slowly, while a newly created client with a fresh session responded almost immediately. After repeated testing, it was finally determined that there was no lag before the context usage reached 25%. However, once it approached around 28-30, every request would result in an abnormally long wait time, with no error messages displayed.

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• Codex model response latency increases significantly as conversation grows #9045: Response latency increases significantly as conversation grows longer (similar symptom of increasing slowness with longer sessions, reported for Codex models)

[ai-llt]

I am using deepseek-chat. The deep thinking mode has not been tested yet.

[ai-llt]

The version I'm using is 1.2.24, and this is a very noticeable issue that has recently been discovered since using the 0.x.x versions.

[ai-llt]

The bottleneck occurs in every conversation and tool invocation. However, there is no lag at all when outputting text content.

[dista]

maybe deepseek server side match context prefix use a lot of time(I guess)?