ModelRegistry.ts•4.96 kB
/**
* Centralized registry of all supported embedding models
* Single source of truth for model information, dimensions, and metadata
*/
export interface ModelInfo {
name: string;
dimensions: number;
provider: "ollama" | "transformers" | "openai";
description: string;
quality: "low" | "medium" | "high" | "highest";
installation?: string;
cost?: string;
}
/**
* Complete registry of all supported embedding models
*/
export const MODEL_REGISTRY: ModelInfo[] = [
// Ollama models (local)
{
name: "mxbai-embed-large",
dimensions: 1024,
provider: "ollama",
description: "High-quality local embeddings, balanced speed and quality",
quality: "high",
installation: "ollama pull mxbai-embed-large",
},
{
name: "nomic-embed-text",
dimensions: 768,
provider: "ollama",
description: "Fast, good quality for general use",
quality: "medium",
installation: "ollama pull nomic-embed-text",
},
{
name: "all-minilm",
dimensions: 384,
provider: "ollama",
description: "Lightweight, fast, lower quality",
quality: "low",
installation: "ollama pull all-minilm",
},
{
name: "snowflake-arctic-embed",
dimensions: 1024,
provider: "ollama",
description: "High-quality, optimized for retrieval tasks",
quality: "high",
installation: "ollama pull snowflake-arctic-embed",
},
// Transformers models (offline)
{
name: "Xenova/all-MiniLM-L6-v2",
dimensions: 384,
provider: "transformers",
description: "Default model, no setup required, downloads on first use",
quality: "low",
},
{
name: "Xenova/all-mpnet-base-v2",
dimensions: 768,
provider: "transformers",
description: "Better quality, larger size, no setup required",
quality: "medium",
},
{
name: "Xenova/bge-small-en-v1.5",
dimensions: 384,
provider: "transformers",
description: "Fast, English-optimized",
quality: "low",
},
{
name: "Xenova/bge-base-en-v1.5",
dimensions: 768,
provider: "transformers",
description: "Better quality, English-optimized",
quality: "medium",
},
// OpenAI models (cloud)
{
name: "text-embedding-3-small",
dimensions: 1536,
provider: "openai",
description: "Cost-effective, high quality cloud embeddings",
quality: "highest",
cost: "$0.020 per 1M tokens",
},
{
name: "text-embedding-3-large",
dimensions: 3072,
provider: "openai",
description: "Best quality, higher cost",
quality: "highest",
cost: "$0.130 per 1M tokens",
},
{
name: "text-embedding-ada-002",
dimensions: 1536,
provider: "openai",
description: "Legacy model, still supported",
quality: "high",
cost: "$0.100 per 1M tokens",
},
];
/**
* Get all models for a specific provider
*/
export function getModelsByProvider(provider: string): ModelInfo[] {
return MODEL_REGISTRY.filter(m => m.provider === provider);
}
/**
* Get model information by name (supports partial matching)
*/
export function getModelInfo(modelName: string): ModelInfo | null {
// Return null for empty string
if (!modelName || modelName.trim().length === 0) {
return null;
}
// Try exact match first
const exactMatch = MODEL_REGISTRY.find(m => m.name === modelName);
if (exactMatch) {
return exactMatch;
}
// Try partial match (for Ollama-style matching)
const partialMatch = MODEL_REGISTRY.find(m =>
modelName.includes(m.name) || m.name.includes(modelName)
);
return partialMatch || null;
}
/**
* Get dimensions for a model by name
* Returns null if model is unknown
*/
export function getModelDimensions(modelName: string): number | null {
const modelInfo = getModelInfo(modelName);
return modelInfo?.dimensions || null;
}
/**
* Get all models from the registry
*/
export function getAllModels(): ModelInfo[] {
return MODEL_REGISTRY;
}
/**
* Check if a model exists in the registry
*/
export function modelExists(modelName: string): boolean {
return getModelInfo(modelName) !== null;
}
/**
* Get models by quality level
*/
export function getModelsByQuality(quality: ModelInfo["quality"]): ModelInfo[] {
return MODEL_REGISTRY.filter(m => m.quality === quality);
}
/**
* Get recommended model for a provider
* Returns the highest quality model available for the provider
*/
export function getRecommendedModel(provider: string): ModelInfo | null {
const models = getModelsByProvider(provider);
// Try to find highest quality first
const highest = models.find(m => m.quality === "highest");
if (highest) {
return highest;
}
// Then try high quality
const high = models.find(m => m.quality === "high");
if (high) {
return high;
}
// Then try medium quality
const medium = models.find(m => m.quality === "medium");
if (medium) {
return medium;
}
// Fall back to first available model (likely "low" quality)
return models[0] || null;
}