Furiosa SDK 2024.2.1 release (Up to 32k context length support)

Announcements
1

We are very excited to announce 2024.2.1 release. 2024.2.1 is the minor release based on 2024.2.0 major release. This release includes a couple of feature improvements, and 32k context length support in models, such as LLaMA 3.1, EXAONE. You can find the upgrade guide for this release in Upgrading Furiosa Software Stack.

Highlights

  • Up to 32k context length (<= 32768) support in furiosa-llm for various models, such as LLaMA 3.1, EXAONE
  • Artifacts with the same tensor_parallelism_size is compatible even with any pipeline_parallel_size

You can find more details about 2024.2.1 release at Release Note of Furiosa SDK 2024.2.1 Beta0.