apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 00:17:25 +08:00
Code Issues Actions 19 Packages Projects Releases Wiki Activity

[BugFix] skip mm revert (#5848)

Browse Source 
* skip mm revert

* update code

* update test
This commit is contained in:
kevin
2026-01-04 14:25:45 +08:00
committed by GitHub
parent e3957a5ebc
commit 52dc9a7b85
4 changed files with 9 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastdeploy/engine/sched/resource_manager_v1.py
+3 -15
View File
@@ -922,21 +922,9 @@ class ResourceManagerV1(ResourceManager):
"""
try:
cache_prepare_time = time.time()
if self._is_mm_request(request) and ErnieArchitectures.is_ernie5_arch(
self.config.model_config.architectures
):
# For multimodal requests using Ernie 5 series models, skip prefix cache.
hit_info = {
"gpu_cache_blocks": 0,
"cpu_cache_blocks": 0,
"gpu_match_token_num": 0,
"cpu_match_token_num": 0,
}
common_block_ids, matched_token_num = [], 0
else:
(common_block_ids, matched_token_num, hit_info) = self.cache_manager.request_match_blocks(
request, self.config.cache_config.block_size
)
(common_block_ids, matched_token_num, hit_info) = self.cache_manager.request_match_blocks(
request, self.config.cache_config.block_size
)
matched_block_num = len(common_block_ids)
no_cache_block_num = self.cache_manager.get_required_block_num(