Skip to content

[Cherry-Pick][BugFix] cp skip_mm_revert(#5848) #5849

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
qingqing01 merged 2 commits into from
Dec 31, 2025
Merged

[Cherry-Pick][BugFix] cp skip_mm_revert(#5848) #5849

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ebef796
cp skip_mm_revert
kevincheng2 Dec 31, 2025
1b345a7
update test
kevincheng2 Dec 31, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 4 additions & 2 deletions fastdeploy/cache_manager/prefix_cache_manager.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1284,8 +1284,10 @@ def _revert_match_blocks(
cpu_match_token_num: int,
swap_node_ids: list,
):
position = request.multimodal_inputs["mm_positions"][chunk_idx]
revert_tokens = matched_token_num - position.offset
# position = request.multimodal_inputs["mm_positions"][chunk_idx]
# revert_tokens = matched_token_num - position.offset
# TODO(chengyanfu): fix when is_chunked_mm_input=True, revert all matched tokens
revert_tokens = matched_token_num
match_block_ids = [node.block_id for node in matche_nodes]
logger.warning(
f"match_block: req_id {request.request_id} revert tokens: {revert_tokens} from matched nodes: {match_block_ids}"
Expand Down
19 changes: 3 additions & 16 deletions fastdeploy/engine/sched/resource_manager_v1.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,6 @@
EncoderCacheManager,
ProcessorCacheManager,
)
from fastdeploy.config import ErnieArchitectures
from fastdeploy.engine.request import (
ImagePosition,
Request,
Expand Down Expand Up @@ -883,21 +882,9 @@ def get_prefix_cached_blocks(self, request: Request):
"""
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(
Expand Down
1 change: 1 addition & 0 deletions tests/v1/cache_manager/test_revert_blocks.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -117,6 +117,7 @@ def test_is_chunked_mm_input_after_last_chunk(self):
self.assertEqual(idx, 0)


@unittest.skip("Skip TestRevertMatchBlocks")
class TestRevertMatchBlocks(unittest.TestCase):
def setUp(self):
self.block_size = 64
Expand Down
Loading