Welcome to Airtrain

Use Airtrain to compare open-source LLMs on your own data, and craft your own domain-specific metrics.

Welcome! Airtrain is a no-code compute platform for language models. It is aimed at AI developers and product builders.

Airtrain focuses on three main compute workloads:

  • Offline batch evaluation of language models. This lets you compare different LLMs on your own datasets, experiment with different configurations and prompts, and define your own domain-specific metrics.
  • Fine-tuning LLMs (coming soon). This lets you specialize pre-trained models to your own tasks and domains.
  • Model serving (coming soon). Serve your fine-tuned models from our affordable scalable infrastructure

With the never-ending stream of new pre-trained Large Language Models coming out almost weekly, it can be difficult to decide which one to use for a specific applications.

Available metrics and benchmarks offer hints of their relative performance, but these are specific to particular use cases and datasets (e.g. summarization, translation, high-school science, etc.) and do not prescribe how a particular model will perform on your specific task.

With Airtrain, you can design evaluation procedures and metrics specific to your application and make better data-driven decisions about your AI stack.

Here are some next steps for you to get started:

What’s Next

Get started with Airtrain.