vllm-project/vllm v0.23.0
DeepSeek-V4 matures across backends: Following its introduction in v0.22.0, DeepSeek-V4 received another large hardening and optimization pass.
Key points
- Please note that Minimax M3 is not yet supported in this version.
- This release features 408 commits from 200 contributors (63 new)!
- Model Runner V2 expands to more dense models: MRv2 is now selected by default for Llama and Mistral dense models (#43458) in addition to Qwen3.
- Transformers v5 compatibility: vLLM now targets Transformers v5, with vendored MiniCPM-V/O processors (#44282) and compatibility fixes for Sarvam (#38804) and Voxtral (#44559).
Sources (1)
- [1]vllm-project/vllm v0.23.0GitHub: vllm-project/vllm · Jun 15, 05:27 AM
* **DeepSeek-V4 matures across backends**: Following its introduction in v0.22.0, DeepSeek-V4 received another large hardening and optimization pass.
Please note that Minimax M3 is not yet supported in this version.
Extractive summary: sentences quoted from the sources.