Back to Ensemble LLMs
Create

Create Ensemble LLM

Configure a single LLM with specific model, parameters, and settings. The resulting configuration can be used directly in your code.

Configuration

The model identifier (e.g., provider/model-name)

Controls randomness (0 = deterministic, 2 = most random)

Nucleus sampling threshold

Enables probabilistic voting (required for vector completions)

Penalizes repeated tokens based on frequency

Penalizes tokens based on presence in text so far

How the LLM should format its output

Amount of reasoning to apply (for supported models)

Comma-separated list of provider preferences

Computed ID

Loading WASM validation...

WASM validation enables real-time ID computation

JSON Configuration

{
  // Enter a model to generate configuration
}

About Ensemble LLMs

  • Content-Addressed: IDs are computed from the definition itself - identical configurations always produce identical IDs
  • Immutable: Once defined, an Ensemble LLM configuration cannot be changed (changing it creates a new ID)
  • No Storage Needed: Copy the JSON and use it directly in your code or save it to GitHub
  • Top Logprobs: Set this (2-20) to enable probabilistic voting for vector completions