Migrating to vLLM v0.25.0
Version v0.25.0 introduces 2 breaking changes. This guide details how to update your code.
Released: 7/11/2026
⚠️ Check Your Code
If you use any of these symbols, you need to read this guide:
PagedAttentionVLLM_GPU_SYNC_CHECKChatOpenAIpydantic.v1torch.compileLLaVA-OneVision-2Unlimited OCRMOSS-Transcribe-Diarizeopenai/privacy-filterHy3GLM-5DeepSeek-V3.2MiniMax-M3Kimi k2.5/k2.6/k2.7seed_ossDeepSeek V4GPTBigCode/Starcoder2RoBERTaVoxtralGemma familyDiffusionGemmaMiniCPM-VWhisperMamba/Mamba2DeepSeek-V2Model Runner V2Speculative decodingDSparkDFlashSleep modeAttentionKV offloadingFlashAttentionFLASH_ATTN_MLA_SPARSEDCPXQANCCLFlashInferMNNVLSequence parallelismData parallelPD disaggregationMooncakeNIXLMultiConnectorROCmIntel XPUCPURISC-VPOWERHelionTritonCuTeDSLFA4-MLAB12xAITERCudaCommunicatorW8A8 FP8 linear kernelAArch64macOS/Apple SiliconOpenMPRVVBF16POWERPCBreaking Changes
●Issue #1
PagedAttention has been removed. Users relying on the legacy attention implementation will need to migrate to the V1/MRv2 backends, which are now the standard path. This change is due to the deprecation and removal of the PagedAttention module.
●Issue #2
The `VLLM_GPU_SYNC_CHECK` environment variable has been introduced. If this variable is set, vLLM will perform GPU synchronization checks, which may impact performance. Users should be aware of this change and its potential performance implications.
Migration Steps
- 1Migrate from PagedAttention to V1/MRv2 backends.
- 2Be aware that the `VLLM_GPU_SYNC_CHECK` environment variable is now available and may impact performance if set.
Release Summary
vLLM v0.25.0 introduces Model Runner V2 as the default for dense models, significantly improving performance and adding support for new features like EVS and realtime embeddings. The release also deprecates PagedAttention and enhances the Transformers backend to match native vLLM speed, alongside numerous model additions and performance optimizations across various hardware platforms.
Need More Details?
View the full release notes and all changes for vLLM v0.25.0.
View Full Changelog