# vLLM v0.2.6 — Release & Upgrade Intelligence
- **Product:** vLLM (https://vllm.ai)
- **Version:** v0.2.6
- **Released:** Dec 17, 2023
- **Impact Score:** 5/10
- **Breaking Changes:** Not verified
- **Canonical Analysis:** https://changelogs.us/vllm/updates/v0.2.6/

## Executive Summary (TL;DR)
vLLM v0.2.6. Source highlight: Fast model execution with CUDA/HIP graph

## New Features
- Add a flag to include stop string in output text
- Add GPTQ support
- [Docs] Add quantization support to docs
- [Docs] Add CUDA graph support to docs
- [Minor] Add more detailed explanation on `quantization` argument
- [Minor] Add Phi 2 to supported models

## Bug Fixes
- Fix memory profiling with tensor parallelism
- Fix *.bin weight loading for Mixtral models
- Fix typing in generate function for AsyncLLMEngine & add toml to requirements-dev
- Fix Dockerfile.rocm
- Fix all-reduce memory usage
- [Minor] Fix xformers version

## Upgrade Command
```bash
python -m pip install 'vllm==0.2.6'
```