●Change8

Migrating to vLLM v0.8.4

Version v0.8.4 introduces 3 breaking changes. This guide details how to update your code.

Released: 4/14/2025

3
Breaking Changes
3
Migration Steps
9
Affected Symbols

⚠️ Check Your Code

If you use any of these symbols, you need to read this guide:

EngineArgsAutoWeightsLoaderLoadConfigParallelConfigPlatform.supports_structured_outputfused_moe_kernelCompressedTensorsW8A8Fp8MoEMethodChatGLMForConditionalGenerationTeleChat2ForCausalLM

Breaking Changes

●Issue #1

The default structured output backend is now set to 'auto' in the V1 engine.

●Issue #2

Multi-input support is now enabled by default in the V1 engine.

●Issue #3

The default 'max_num_seqs' for the V1 engine has been reverted to V0 values for most hardware platforms.

Migration Steps

  1. 1
    If using Llama4, update immediately to apply critical accuracy fixes.
  2. 2
    Review custom structured output configurations as the default backend is now 'auto'.
  3. 3
    Check memory settings as max-model-len estimation now uses available KV cache memory.

Release Summary

This release introduces support for Llama4 and Qwen3 models, alongside significant performance optimizations for DeepSeek MLA and MoE kernels. It also stabilizes the V1 engine by enabling multi-input and structured outputs by default.

Need More Details?

View the full release notes and all changes for vLLM v0.8.4.

View Full Changelog