Release Notes#

Release 2.1.0#

Summary#

This release introduces support for the Evo 2 7B 1M model variant, providing a faster and more resource-efficient option while maintaining strong accuracy for most use cases. The 40B 1M model remains available as the default for maximum accuracy. This update also includes documentation improvements, clarifies advanced usage and model selection, and updates the release notes to reflect the new model variant.

Since the container now supports both 40B/7B variants and custom models, the name of the container has changed from evo2-40b to just evo2. New releases are still published under both names for compatibility and convenience of existing users.

Model Variants#

  • Evo 2 7B 1M

  • Evo 2 40B 1M

New features and improvements#

  • Added support for the Evo 2 7B 1M model variant, selectable via the NIM_VARIANT=7b environment variable.

  • Updated documentation to describe model selection and advanced usage.

  • Container name changed from evo2-40b to evo2 to reflect support for multiple model variants and custom models.

Bug fixes#

  • Minor documentation corrections and formatting fixes.

Performance enhancements#

  • The 7B model variant offers reduced memory and compute requirements, enabling faster inference.

Breaking changes#

  • None.

Migration notes#

  • No migration steps are required for existing users of the 40B model. To use the 7B model, set the NIM_VARIANT environment variable to 7b when launching the container.

  • The container is now available under the name evo2 instead of evo2-40b, though both names remain available for compatibility.

Release 2.0.0#

Summary#

This major release introduces support for loading custom and fine-tuned checkpoints, enabling users to leverage their own model weights or community-provided variants. The internal implementation of NIM has migrated to the NVIDIA BioNeMo Framework, providing unified training and inference code bases and standardized checkpoint formats.

Thanks to the migration to the BioNeMo Framework, new Evo 2 NIM release delivers up to 30% faster inference workflows compared to the 1.x series.

Important: Due to the change in internal model representation, model layer names and parameter shapes have changed. As a result, /forward API calls are not backward-compatible with 1.x releases.

Model Variant#

  • Evo 2 40B 1M

Release 1.0.2#

Summary#

This is a minor update that includes changes that enhance stability for users operating in multi-GPU environments and improve compatibility with new NIM versions:

  • Resolved an issue related to FP8 extra states when running on specific combinations of GPU interconnects and CPUs. The fix ensures correct handling of FP8 extra states, with additional checks to verify correct functionality.

  • Updated nimlib and related dependencies to improve compatibility with the latest NIM releases.

Model Variant#

  • Evo 2 40B 1M

Release 1.0.1#

Summary#

This is a minor update that includes the following changes:

  • The input prompt is now automatically converted to uppercase internally, helping to prevent the common mistake of supplying lowercase DNA sequences to the model.

  • For non-recoverable errors, such as unrecoverable memory ECC errors, the health endpoint will now reflect the error status. This allows cloud infrastructure to redeploy the instance to healthy hardware.

Model Variant#

  • Evo 2 40B 1M

Release 1.0.0#

Summary#

This is the first release of Evo 2 NIM. Evo 2 is a biological foundation model that is able to integrate information over long genomic sequences, while retaining sensitivity to single-nucleotide changes.

This release features 40 billion parameters model trained on 11 trillion DNA nucleotides with context-length of up to 1 million DNA nucleotides.

Model Variant#

  • Evo 2 40B 1M

Notes and Limitations#

Ensure you use this model with GPUs that have FP8 support, such as NVIDIA H100, NVIDIA H200, or newer. For more information, see Prerequisites and Support Matrix for Evo2 NIM. Older GPUs are not currently supported.

Ensure you always use the benchmarking suite so that the deployed model works as expected.