# vLLM v0.8.2 — Release & Upgrade Intelligence
- **Product:** vLLM (https://vllm.ai)
- **Version:** v0.8.2
- **Released:** Mar 23, 2025
- **Impact Score:** 5/10
- **Breaking Changes:** Not verified
- **Canonical Analysis:** https://changelogs.us/vllm/updates/v0.8.2/

## Executive Summary (TL;DR)
vLLM v0.8.2. Source highlight: This release contains important bug fix for the V1 engine's memory usage. We highly recommend you upgrading!

## New Features
- Add flag to disable cascade attention (#15243)
- Scheduler Refactoring: Add Scheduler Interface (#15250)
- Add `disable-any-whitespace` option support for xgrammar (#15316)
- Integrate `fastsafetensors` loader for loading model weights (#10647)
- Add guidance backend for structured output (#14589)
- Add Kubernetes deployment guide with CPUs (#14865)

## Improvements
- [FrontEnd][Perf] `merge_async_iterators` fast-path for single-prompt requests
- [Model] RE: Mamba2 Prefill Performance Tweaks: Fixing Flurry of Unnecessary Memory Copies
- [V1][Perf] Simpler request output queues

## Bug Fixes
- Fix V1 Engine crash while handling requests with duplicate request id (#15043)
- [BugFix][V1] Update stats.py
- [BugFix] Lazily import XgrammarBackend to avoid early cuda init
- Fix broken tests
- [Bugfix] Fix embedding assignment for InternVL-based models
- fix "Total generated tokens:" is 0 if using --backend tgi and --endpo…

## Upgrade Command
```bash
python -m pip install 'vllm==0.8.2'
```