Morph MORPH
SIGN IN SIGN UP
tejas / vllm

A high-throughput and memory-efficient inference and serving engine for LLMs

0 0 122 Python

[Bugfix] Use PIECEWISE cudagraphs on Blackwell if max_model_len > 131072 (#27114)

Signed-off-by: mgoin <mgoin64@gmail.com>
M
Michael Goin committed
950cf9e58eefd8e6436eb04b442184353c06a571
Parent: 3125d79
Committed by GitHub <noreply@github.com> on 10/17/2025, 7:47:18 PM