Introduction

Introduction#

Why Materialized?#

It can be incredibly useful to use large, generalized AI models to process datasets in new and interesting ways. From sentiment analysis, to structured data extraction, to classification, to complex reasoning and more - LLM’s are just being recognized as a new means of working with data.

However, working with LLM’s in such scenarios can be a pain, and there are few tools for doing so currently. Between managing GPU infrastructure, tuning inference frameworks, calculating/managing costs, caching results, evaluating quality, and creating task-specific fine-tunes, it quickly becomes painful to iterate towards a production-grade solution.

We aim to solve these issues by providing a platform for large-scale, batch and offline inference with LLM’s. We handle the infrastructure, inference optimizations, and cost management, so you can focus on solving the problems you care about.

With Materialized, you can expect improvements in:

  • Cost: you can expect to pay significantly less than real-time inference providers.

  • Speed: you can expect large-scale jobs (thousands of requests or more) to finish in minutes, allowing you to iterate quickly.

  • Quality: you can use our tooling to rapidly iterate towards a production-grade solution, customized to your specific needs.

  • Ease: you can use our SDKs and CLI to quickly get up and running, without having to worry about the underlying infrastructure.

Ready to get started? Refer to the Batch API Reference for more details, or check out our Quickstart page to get up and running immediately.

When to use Materialized#

You’re using LLM’s for offline, large-scale, analytical workloads or data transformations - likely over tables or collections of objects. Latency isn’t nearly as important as cost or quality.

When not to used Materialized#

You’re serving a user-facing application or product with one-off, online inference calls (e.g. a chatbot) and latency is critical. For such use-cases, we recommend using another inference provider who optimizes for latency.

Not sure if Materialized is right for you, or not sure how to get started?#

We’d love to hear about your use case, and help you figure out if we offer the right solution. We also offer bespoke, custom solutions for enterprise customers. Please contact us at team@materialized.dev.

Contents