dsh-gray-test: Gray Model Detection Tool

`dsh-gray-test` is a DSH plugin maintained by LiLiLi-Zi that detects whether a local model is the official version or a gray-scale debugging version. It performs its determination based on specific sentence-pattern counting rules by running concurrent tests and streaming the chain-of-thought. The plugin supports one-click concurrent testing (1–100 sessions) and real-time monitoring and analysis of the chain-of-thought content. The detection logic prioritizes gray-scale features: if the count of `I'm/I'll + verb-ing` is greater than five times the count of `Let me`, the model is classified as gray-scale; otherwise, if `Let me` appears at least 3 times, it is classified as normal. Test results are logged in the `~/gray-test-runs/` directory. Users must configure the API Key for `deepseek-official` in DSH settings before running the plugin.

Read More