vLLM Releases v0.33.0 With New Models and Speculative Decoding Updates

What happened

vLLM has released v0.33.0, a version built from 927 commits with contributions from 389 contributors, including 122 first-timers. It adds support for Reflection Beam-1.0, Step-5 Preview, Nemotron Parse 3.5, and MiniCPM-OCR 1.0, and includes PyTorch 2.14 wheels that require CUDA 13.0 and are libtorch ABI-stable. Speculative decoding updates include the XPress refiner and FLy lossy verify.

Read next

• More changes in today’s briefingHugging Face Proposes Model Integration Contracts for Transformers

The story so far

vLLM released version 0.33.0 with support for Reflection Beam-1.0, Step-5 Preview, Nemotron Parse 3.5, and MiniCPM-OCR 1.0. The release also adds the XPress refiner and FLy lossy verify for speculative decoding. Its PyTorch 2.14 wheels require CUDA 13.0 and are libtorch ABI-stable. vLLM said the release includes 927 commits from 389 contributors, including 122 first-time contributors.