vLLM says shipping precompiled Python bytecode in its images cuts startup time 14–18%, about 18 seconds on an H100. The change is a single uv flag at build time, so containers no longer recompile thousands of Python files at launch. It is in CUDA nightlies and in official release images from v0.32.0.
Cold start, meet build time. 📷
Shipping precompiled Python bytecode in our images cuts vLLM startup by 14–18% (~18s on H100) — one uv flag at build time, zero recompiles at launch.
Already in CUDA nightlies, official release images from v0.32.0. Great work @nilsmatteson 📷
Fresh vLLM containers can recompile thousands of Python files at startup. One build flag moves that work to image build time: 14–18% less startup time (~18 s) in my H100 benchmarks. Already in the CUDA nightly; official release images from v0.32.0. https://vllm.ai/blog/2026-09-22-python-bytecode-startupView quoted post on X
Source: vLLM · x.comPublished · added here
