# vLLM v0.7.3 — Release & Upgrade Intelligence
- **Product:** vLLM (https://vllm.ai)
- **Version:** v0.7.3
- **Released:** Feb 20, 2025
- **Impact Score:** 5/10
- **Breaking Changes:** Not verified
- **Canonical Analysis:** https://changelogs.us/vllm/updates/v0.7.3/

## Executive Summary (TL;DR)
vLLM v0.7.3. Source highlight: Deepseek enhancements:

## New Features
- Add intial ROCm support to V1 (#12790)
- Add sleep and wake up endpoint and v1 support (#12987)
- Add `/v1/audio/transcriptions` OpenAI API endpoint (#12909)
- [Docs] Add Google Cloud Slides
- Add Bamba Model
- [Model] Add support for partial rotary embeddings in Phi3 model

## Improvements
- Deepseek enhancements:
- Metrics enhancements: GPU prefix cache hit rate % gauge (#12592), iteration_tokens_total histogram (#13288),  several request timing histograms (#12644)
- Enhancement to Qwen2.5-VL: BNB support (#12944), LoRA (#13261), Optimizations (#13155)
- Neuron: Support Longer Sequences in NKI-based Flash PagedAttention and Improve Efficiency (#12921)
- Reduce TTFT with concurrent partial prefills (#10235)
- LoRA - Refactor sgmv kernels (#13110)

## Bug Fixes
- Fixes for cases of FA2 illegal memory access error (#12848)
- [Bugfix] Fix unsupported FA version check for Turing GPU
- [Bugfix] FA2 illegal memory access
- [Bugfix] Missing quant_config in deepseek embedding layer
- [ROCm] [Feature] [Doc] [Dockerfile] [BugFix] Support Per-Token-Activation Per-Channel-Weight FP8 Quantization Inferencing
- [Bugfix] Fix disagg hang caused by the prefill and decode communication issues

## Upgrade Command
```bash
python -m pip install 'vllm==0.7.3'
```