Introduction

In model inference scenarios, estimating the concurrent request volume (QPS) is a key step in assessing system capacity. This task usually requires external scripts or manual calculation formulas.

DeepSeek Harness (DSH) adopts the “everything is a plugin” philosophy. To address this need, the community provides a plugin named dsh-qps-calc, which allows you to calculate QPS directly within a session.

Plugin Positioning

This is a DeepSeek Harness plugin maintained by uckkk, classified under the model inference category. Its core purpose is to provide peak QPS estimation, helping developers quickly assess the concurrent capacity of inference services.

Core Features

  • qps_calc: Accurately calculates peak QPS based on parameters and returns detailed results.
  • Implementation: Pure Node implementation with no network dependencies.

Installation and Dependencies

Run the following command to install the plugin:

dsh plugin add github:uckkk/dsh-qps-calc

Before installation, ensure the current environment meets the following dependency requirements:

  • @deepseek-ai/cordis (^4.0.1)
  • @deepseek-ai/dsh-tools (>=0.1.0-rc.6)

Usage

After installation, simply call the tools registered by this plugin in a DSH session.

Use Cases and Notes

  • Suitable scenarios: Suitable for developers who need to quickly evaluate the peak concurrent volume of inference services locally.
  • Permissions: The plugin runs with the permissions of the current DSH process.
  • Security advice: It is recommended to review the source code and license (MIT) before installation.
  • Ecosystem note: This plugin belongs to the community ecosystem. The community catalog (skillhub.cn) has no official affiliation with DeepSeek or High-Flyer.

Conclusion

dsh-qps-calc provides a lightweight QPS estimation solution with no network dependencies. For more details, refer to the GitHub repository or the community catalog.