# vLLM v0.6.0 — Release & Upgrade Intelligence
- **Product:** vLLM (https://vllm.ai)
- **Version:** v0.6.0
- **Released:** Sep 04, 2024
- **Impact Score:** 5/10
- **Breaking Changes:** Not verified
- **Canonical Analysis:** https://changelogs.us/vllm/updates/v0.6.0/

## Executive Summary (TL;DR)
vLLM v0.6.0. Source highlight: We are excited to announce a faster vLLM delivering 2x more throughput compared to v0.5.3. The default parameters should achieve great speed up, but we recommend also try out turning on multi step scheduling. You can do so by setting `--num-scheduler-steps 8` in the engine arguments. Please note tha

## New Features
- OpenAI-Compatible Tools API + Streaming for Hermes & Mistral models! (#5649)
- Add json_schema support from OpenAI protocol (#7654)
- Enable chunked prefill and prefix caching together (#7753, #8120)
- Multimodal support in offline chat (#8098), and multiple multi-modal items in the OpenAI frontend (#8049)
- [Core] Add multi-step support to LLMEngine
- [Frontend] add json_schema support from OpenAI protocol

## Improvements
- We are excited to announce a faster vLLM delivering 2x more throughput compared to v0.5.3. The default parameters should achieve great speed up, but we recommend also try out turning on multi step scheduling. You can do so by setting `--num-scheduler-steps 8` in the engine arguments. Please note that it still have some limitations and being actively hardened, see #7528 for known issues.
- Multi-step scheduler now supports LLMEngine and log_probs (#7789, #7652)
- Asynchronous output processor overlaps the output data structures construction with GPU works, delivering 12% throughput increase. (#7049, #7911, #7921, #8050)
- Using FlashInfer backend for FP8 KV Cache (#7798, #7985), rejection sampling in Speculative Decoding (#7244)
- [misc][cuda] improve pynvml warning
- [Performance][BlockManagerV2] Mark prefix cache block as computed after schedule

## Bug Fixes
- [Bugfix] Fix run_batch logger
- [ci][test] fix RemoteOpenAIServer
- [Bugfix] Fix Phi-3v crash when input images are of certain sizes
- [Bugfix]: Use float32 for base64 embedding
- [misc] fix custom allreduce p2p cache file generation
- [Bugfix] neuron: enable tensor parallelism

## Upgrade Command
```bash
python -m pip install 'vllm==0.6.0'
```