custom provider · Model inference strength setting
534119219/dsh-custom-provider-reasoning · model-adapters
dsh plug-in: allows all models of the custom provider (pi-ai handwritten routing) to select the thinking intensity (inference level) and choose to actually send them online through the original adapter.
★ 4GitHub stars
0Forks
2026-08-17Last updated
JavaScriptLanguage
MITLicense
Key features
- Composer's model selector only displays the reasoning level line if the adapter publishes reasoning metadata for that exact model
- openai-completionsRoutes whose model id contains deepseek are automatically completed compat.thinkingFormat: deepseek (use DeepSeek's thinking parameter dialect instead of OpenAI's reasoning_effort).
- Endpoint compatibilityOpenAI compatible reasoning_effort vocabulary injection by default.
- planOps() is a pure function that calculates the minimum settings.mutate path operation.
- dsh-llm-pi-ai will only materialize this metadata from the reasoningEfforts statement of the model entry in the llm-pi-ai settings section
Install command
dsh plugin --profile web add github:534119219/dsh-custom-provider-reasoning