# vLLM v0.8.4 — Release & Upgrade Intelligence
- **Product:** vLLM (https://vllm.ai)
- **Version:** v0.8.4
- **Released:** Apr 14, 2025
- **Impact Score:** 5/10
- **Breaking Changes:** Not verified
- **Canonical Analysis:** https://changelogs.us/vllm/updates/v0.8.4/

## Executive Summary (TL;DR)
vLLM v0.8.4. Source highlight: This release includes important accuracy fixes for Llama4 models, if you are using it, we highly recommend you to update.

## New Features
- Add tuned FusedMoE kernel config for Llama4 Scout, TP=8 on H100  (#16488)
- Added chat templates for LLaMa4 pythonic tool calling (#16463)
- Add warning for Attention backends that do not support irope yet (#16212)
- Add hf_token to EngineArgs (#16093)
- Add bucket for `request_latency`, `time_to_first_token` and `time_per_output_token` (#15202)
- Add support to modelopt quantization of Mixtral model (#15961)

## Improvements
- DeepSeek MLA: a new merge_attn_states CUDA kernel, 3x speedup (#16173)
- MoE: Support W8A8 channel-wise weights and per-token activations in triton fused_moe_kernel (#16366)
- Enable PTPC FP8 for CompressedTensorsW8A8Fp8MoEMethod (triton fused_moe) (#16537)
- [Misc] Improve model redirect to accept json dictionary
- [Misc] improve example mlpspeculator and llm_engine_example
- Improve configs - `ParallelConfig`

## Bug Fixes
- Llama4 (#16113,#16509) bug fix and enhancements:
- [CI] Fix benchmark script level
- fix: support clang17 for macos and fix the real libomp
- [doc] fix 404
- Fix some capitalisations in generated examples doc titles
- [Bugfix] fix use_atomic_add support of marlin kernel when using v1 engine

## Upgrade Command
```bash
python -m pip install 'vllm==0.8.4'
```