reka-ai/vllm-reka v0.17.0

v0.17.0

Repository: reka-ai/vllm-reka

Tag: v0.17.0

Published: 2026-04-13T09:45:19Z

Prerelease: no

Release notes:

Compatibility

This release is compatible with vLLM 0.17.x (dependency pin: vllm>=0.17.0,<0.18.0).

Also usable alongside vllm-omni, which depends on vLLM 0.17.x.

For vLLM 0.15.x, use v0.15.0. There is no release targeting 0.16.

What changed

Internal updates to track vLLM API changes between 0.15 and 0.17:

• Attention import moved to vllm.model_executor.layers.attention
• Added get_language_model() on multimodal model classes (required by the new SupportsMultiModal default embed_input_ids)
• _get_data_parser moved from processor to BaseProcessingInfo.get_data_parser
• ProcessorInputs renamed mm_data → mm_data_items (now takes MultiModalDataItems, not MultiModalDataDict)

Install

pip install git+https://github.com/reka-ai/vllm-reka.git@v0.17.0