Configuration
Learn how to configure model, provider, and generation parameters for your prompts.
The configuration section, located at the top of every prompt file and enclosed by triple dashes (---
), defines how Latitude executes your prompt. It’s written in YAML format and allows you to specify the AI provider, model, generation parameters, and other advanced settings.
Core Settings
Provider (required)
Specifies the AI provider to use (e.g., OpenAI, Anthropic, Google). This must match a provider configured in your Latitude workspace settings.
You can easily select a configured provider using the dropdown in the editor’s header.
Learn more about configuring providers.
Model (required)
Specifies the exact language model to use (e.g., gpt-4o-mini
, claude-3-opus-20240229
). Available models depend on the selected provider.
The model dropdown in the editor’s header lists available models for the chosen provider.
Generation Parameters
These parameters control how the AI model generates its response. The specific ranges and behaviors might vary slightly between providers.
temperature
Controls the randomness of the output. Lower values (e.g., 0.1) make the output more deterministic and focused, while higher values (e.g., 0.9) increase creativity and randomness.
Setting temperature to 0, or leaving it unset, enables response caching for identical inputs.
It’s generally recommended to adjust either temperature
or topP
, but not both.
maxTokens
Sets the maximum number of tokens (words or parts of words) the model can generate in its response.
topP
(Nucleus Sampling)
An alternative to temperature
for controlling randomness. It instructs the model to consider only the most probable tokens whose cumulative probability mass exceeds the topP
value (e.g., 0.9 means consider tokens comprising the top 90% probability mass).
topK
Restricts the model to sampling only from the K
most likely next tokens at each step. Generally used for advanced cases; temperature
or topP
are usually sufficient.
presencePenalty
Discourages the model from repeating information already present in the prompt context. Higher values increase the penalty.
frequencyPenalty
Discourages the model from using the same words or phrases repeatedly in its response. Higher values increase the penalty.
stopSequences
A list of strings that, if generated by the model, will cause generation to stop immediately.
seed
An integer used to initialize the random number generator. If supported by the model, using the same seed with identical inputs will produce deterministic outputs, useful for reproducibility.
Advanced Configuration
parameters
Defines types and constraints for input parameters used in the Playground or shared prompts.
Learn more about parameter types in the Playground guide.
schema
Defines a JSON schema to enforce structured output from the model.
Learn more about enforcing JSON output.
tools
Lists the tools (functions) available for the AI model to call.
Learn more about configuring tools.
maxSteps
For Chains and Agents, sets the maximum number of execution steps allowed (default: 20, max: 150).
maxRetries
Maximum number of times to retry a provider call on failure (default: 2).
headers
Sends additional HTTP headers with the request, useful for integrating with specific provider features or observability tools.
Next Steps
- Test prompts with different settings in the Playground
- Learn about Enforcing JSON Output
- Explore using Tools and Agents