A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[Bugfix] Fix incompatibility between #20452 and #24548 (#24754)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Sep 13, 2025 at 02:17 UTC
010acc6e1ea1d59ef530459e1078e8fde80f2082
1 file changed
+2
-4
vllm/v1/executor/utils.py
+2
-4
@@ -19,7 +19,5 @@ def get_and_update_mm_cache(
19
"""
20
scheduler_output = args[0]
21
for request_data in scheduler_output.scheduled_new_reqs:
22
- for i in range(len(request_data.mm_kwargs)):
23
- mm_input = request_data.mm_kwargs[i]
24
- request_data.mm_kwargs[i] = \
25
- receiver_cache.get_and_update_item(mm_input, None)
22
+ request_data.mm_features = receiver_cache.get_and_update_features(
23
+ request_data.mm_features)