Add and Configure Custom FastGPT Models

Add and Configure Custom FastGPT Models

Overview

If FastGPT's built-in models do not meet your deployment requirements, you can add custom model configurations. If the submitted Model ID matches an existing built-in model ID, the entry will modify the existing model rather than creating a new one. Two configuration methods are available: web-based admin form entry, or direct configuration file editing. The file method is ideal for replicating consistent model setups across multiple FastGPT instances, as it enables quick bulk configuration.

Configuration Methods

Web Form Setup

Use the FastGPT admin interface's model configuration form to add custom models. The form will automatically prompt for relevant metadata based on your selected model category. Two sample form screenshots are provided in the official documentation to guide setup.

Configuration File Setup

For automated or bulk configuration, edit the model configuration files directly. Each custom model entry follows a standardized JSON structure, with fields tailored to the model's functional category. All custom model entries share these base fields:

Model-Specific Fields

• Language Models: Add maxContext, maxResponse, quoteMaxToken, maxTemperature, charsPointsPrice, censor, vision, toolChoice, functionCall, customCQPrompt, customExtractPrompt, defaultSystemChatPrompt, defaultConfig, and fieldMap.
• Embedding Models: Include defaultToken (default text split token count) and maxToken (maximum allowed token count).
• Rerank Models: Require type set to "rerank", plus optional requestUrl and requestAuth for custom API endpoints.
• Text-to-Speech Models: Require type set to "tts", a voices array of voice options with label and value fields, and charsPointsPrice.
• Speech-to-Text Models: Use the core base schema with no additional type-specific required fields.

Source: FastGPT official documentation and source