How many dimensions should I use for RAG?

While the default is 3072, this skill recommends 768 dimensions for most RAG applications to balance storage costs and performance without significantly sacrificing accuracy.

What is the benefit of specifying a taskType?

Using task types like RETRIEVAL_QUERY or RETRIEVAL_DOCUMENT optimizes the embedding for its specific role, typically improving search relevance by 10-30%.

Can I use this in serverless environments like Cloudflare Workers?

Absolutely. It includes specific patterns for using the standard Fetch API for environments where the full Node.js SDK might not be supported.

What model does this skill support?

This skill is optimized for the production-ready gemini-embedding-001 model, providing the most stable and reliable performance for production apps.

Does this handle the Gemini API rate limits?

Yes, the skill includes implementation patterns for exponential backoff and batch chunking to handle both Free and Tier 1 rate limits effectively.

Google Gemini Embeddings

Name: Google Gemini Embeddings
Author: secondsky

bysecondsky

•

Data Science & ML

Implements Google Gemini embeddings API for high-performance RAG and semantic search applications.

About

This skill streamlines the integration of Google's gemini-embedding-001 model into your development workflow, providing standardized patterns for generating high-quality vector embeddings. It covers essential production requirements including Matryoshka representation learning for flexible dimensions (768 to 3072), task-specific optimization for queries versus documents, and robust error handling for rate limits and text truncation. Ideal for developers building RAG systems, semantic search engines, or document clustering tools, it ensures implementation consistency across different environments like Node.js and Cloudflare Workers.

Key Features

Cloudflare Workers and Vectorize integration patterns
Automatic handling of the 2,048 token input limit via chunking
21 GitHub stars
Specialized task types for optimized search and classification results
Flexible dimensionality (768, 1536, 3072) using Matryoshka learning
Production-ready batch processing with exponential backoff

Use Cases

Automating document clustering and topic modeling
Building Retrieval-Augmented Generation (RAG) pipelines
Implementing semantic search and similarity matching

About

Key Features

Cloudflare Workers and Vectorize integration patterns
Automatic handling of the 2,048 token input limit via chunking
21 GitHub stars
Specialized task types for optimized search and classification results
Flexible dimensionality (768, 1536, 3072) using Matryoshka learning
Production-ready batch processing with exponential backoff

Use Cases

Automating document clustering and topic modeling
Building Retrieval-Augmented Generation (RAG) pipelines
Implementing semantic search and similarity matching