Skip to content
All flagships

Experimental foundation checkpoint

Eve-3

Study a custom architecture through an open checkpoint.

The public 0.5B SABER checkpoint is a foundation-stage model for architecture experiments. Its files, custom modeling code, and training-stage notes are available for inspection.

Inspect the 0.5B checkpoint Follow the guide

What you can do with it

Who it's for: Researchers comparing custom transformer components and training behavior.

A precisely identified checkpoint for a controlled architecture or generation experiment.

Your first useful result

  1. 01

    Identify the artifact

    Use anthonym21/eve-3-0.5b-saber for the public 0.5B checkpoint. Inspect its config, model code, and foundation-stage training note.

  2. 02

    Follow the checkpoint loading example

    The Hugging Face card provides the AutoModelForCausalLM and AutoTokenizer loading calls. Keep the exact model ID when following them.

  3. 03

    Define one controlled test

    Record a small prompt set, generation settings, and a baseline checkpoint. Compare results before attributing an effect to the architecture.

Know the limits

  • The public card describes a foundation checkpoint at 10.8B of a planned 50B training tokens. It does not establish instruction-following quality.
  • The GitHub repository documents a separate 1B design. Its architecture claims and loading example should not be assumed to describe the public 0.5B artifact.

Source & further reading