# vLLM v0.5.2 — Release & Upgrade Intelligence
- **Product:** vLLM (https://vllm.ai)
- **Version:** v0.5.2
- **Released:** Jul 15, 2024
- **Impact Score:** 7/10
- **Breaking Changes:** Source flagged
- **Canonical Analysis:** https://changelogs.us/vllm/updates/v0.5.2/

## Executive Summary (TL;DR)
vLLM v0.5.2. Source highlight: ❗Planned breaking change ❗: we plan to remove beam search (see more in #6226) in the next few releases. This release come with a warning when beam search is enabled for the request. Please voice your concern in the RFC if you do have a valid use case for beam search in vLLM

## Breaking Changes & Deprecations
- ❗Planned breaking change ❗: we plan to remove beam search (see more in #6226) in the next few releases. This release come with a warning when beam search is enabled for the request. Please voice your concern in the RFC if you do have a valid use case for beam search in vLLM

## New Features
- Add PaliGemma (#5189), Fuyu-8B (#3924)
- Enabling bonus token in speculative decoding for KV cache based models (#5765)
- Medusa Implementation with Top-1 proposer (#4978)
- An experimental vLLM CLI for serving and querying OpenAI compatible server (#5090)
- Add support for multi-node on CI (#5955)
- Benchmark: add H100 suite (#6047)

## Improvements
- ZeroMQ fallback for broadcasting large objects (#6183)
- Simplify code to support pipeline parallel (#6406)
- Turn off CUTLASS scaled_mm for Ada Lovelace (#6384)
- Use CUTLASS kernels for the FP8 layers with Bias (#6270)
- [Doc] Move guide for multimodal model and other improvements
- @ericperfect made their first contribution in https://github.com/vllm-project/vllm/pull/6203

## Bug Fixes
- Update wheel builds to strip debug
- Fix release wheel build env var
- add benchmark for fix length input and output
- [Bugfix] FIx benchmark args for randomly sampled dataset
- [Bugfix] use diskcache in outlines _get_guide  #5436
- [Bugfix] Mamba cache Cuda Graph padding

## Upgrade Command
```bash
python -m pip install 'vllm==0.5.2'
```