infos

The infosmethod, when used with Polyfiregenerate function, lets you see more details about how the text is created.

Infos Option

Use the infos method to get a report showing token use (what goes in and comes out) and the resources the model uses when creating text with the web, memory or data options. This is helpful for seeing how the model works, its costs, and figuring out where the created content comes from.

To use the infos feature, call the infos method of the result of the generate function:

TypeScript

generate("your_task_here").infos();

When turned on, the generate function gives back a GenerationResult that includes:

type GenerationResult = {
    result: string;  // The generated content
    tokenUsage: {    // Token usage for input and output
        input: number;
        output: number;
    };
    ressources?: Ressource[];  // Resources or embeddings used during generation
};

TokenUsage

A part of GenerationResult, TokenUsage shows clearly the tokens used in a specific text creation task:

TypeScript

type TokenUsage = {
    input: number;  // Tokens consumed for the input task.
    output: number; // Tokens generated as output.
};

This feature is especially valuable for understanding model computation costs and ensuring you're not overshooting token limitations.

Ressource

Ressource arrays encapsulate and present the various resources from web, memory or data that the model relied upon during generation:

TypeScript

type Ressource = {
    similarity: number;  // Similarity score between resource and generated content.
    id: string;          // Unique identifier of the resource.
    content: string;     // Actual content of the resource.
};

This data can be leveraged to trace back the origins of generated content.

💡 Example

TypeScript

// In React:
const { models: { generate } } = usePolyfire();
// In other environments:
const { models: { generate } } = polyfire;

// Define task, options, and client configurations
const task = "Explain the concept of quantum entanglement.";

const generationOptions: GenerationWithInfosOptions = {
  provider: "openai",
  model: "gpt-3.5-turbo",
  memoryId: "your_memory_id",
};

try {
  const result = await generate(task, generationOptions).infos();
  console.log("Generated Output:", result.result);
  console.log("Token Usage:", result.tokenUsage);
  console.log("Resources Used:", result.ressources);

} catch (error) {
  console.error("Generation faced an inconsistency:", error);
}