LLM Import

Common questions this page answers:

• How do I enable LLM-powered model card import?
• What LLM providers are supported?
• What configuration options are available?
• How do users interact with the import feature?

Bailo can optionally use a large language model (LLM) to extract model card data from existing written documentation. This page explains how to enable and configure the feature. For how users run an import, see Importing a Model Card from Text.

Configuring LLM import

LLM import requires an OpenAI-compatible chat completions endpoint. Configure it under and set ui.llmImport.enabled to true:

{
  llm: {
    endpoint: 'https://api.openai.com/v1/chat/completions',
    apiKey: 'your-api-key',
    model: 'gpt-4o',
    maxTokens: 16384,
    timeoutMs: 120000,
    temperature: 0,
  },

  ui: {
    llmImport: {
      enabled: true,
    },
  },
}

For Helm deployments, set these values in values.yaml under llm and config.ui.llmImport. The API key is read from the <release>-llm-api secret (key auth-token) rather than values.yaml. See App Configuration for how configuration overrides are applied.

Configuration options

• ui.llmImport.enabled - Shows the Import from Text button in the model card editor. When false, the button is hidden and the backend rejects import requests. Default: false
• llm.endpoint - URL of an OpenAI-compatible chat completions endpoint
• llm.apiKey - API key for authenticating with the LLM provider
• llm.model - Model identifier for the LLM provider (for example gpt-4o or gpt-4.1-mini)
• llm.maxTokens - Maximum number of tokens in the LLM response. Default: 16384
• llm.timeoutMs - Timeout for LLM requests in milliseconds. Default: 120000
• llm.temperature - Controls response determinism. A low value produces more deterministic responses. Default: 0
• llm.systemPrompt - System prompt sent to the LLM. The default prompt instructs the model to extract only explicitly stated information and avoid generating placeholder values. You may want to tune this for your specific LLM and schema

How it works

When a user runs an import, Bailo sends their text and the model card schema to the configured LLM, using the llm.systemPrompt instructions. The LLM response is validated against the schema before it reaches the model card form. Fields that do not match the schema or contain placeholder values are removed, and users see these as validation warnings.

Pasted text is sent to the configured endpoint, so choose an LLM provider that is approved for the documentation your users will import.

Related pages

• Importing a Model Card from Text - Using the feature (for users)
• App Configuration - General application configuration