From 3da24c2df735354ccb463650c29cca8ce506fa07 Mon Sep 17 00:00:00 2001 From: Caio Mendes Date: Sat, 27 Apr 2024 07:08:15 -0300 Subject: [PATCH] [Model] Phi-3 4k sliding window temp. fix (#4380) --- vllm/core/block_manager_v1.py | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/vllm/core/block_manager_v1.py b/vllm/core/block_manager_v1.py index be093922b84f2..1fac2636e86fa 100644 --- a/vllm/core/block_manager_v1.py +++ b/vllm/core/block_manager_v1.py @@ -1,4 +1,5 @@ """A block manager that manages token blocks.""" +import math from abc import ABC, abstractmethod from itertools import count, takewhile from os.path import commonprefix @@ -220,9 +221,9 @@ class BlockSpaceManagerV1(BlockSpaceManager): self.block_sliding_window = None if sliding_window is not None: - assert sliding_window % block_size == 0, (sliding_window, - block_size) - self.block_sliding_window = sliding_window // block_size + # Round up to nearest block size to regularize sliding window + # allocation sizes. + self.block_sliding_window = math.ceil(sliding_window / block_size) self.watermark = watermark assert watermark >= 0.0