[Feat] Asu store adaptation#1018
Merged
mag1c-h merged 6 commits intoJun 15, 2026
Merged
Conversation
Fengli5355
requested review from
FangRun2,
Infinite666,
Tarrei,
Wwwzff,
harrisonyhq,
mag1c-h,
nrj868,
qyh111 and
ygwpz
as code owners
Infinite666
reviewed
Jun 11, 2026
Fengli5355
force-pushed
the
asu-store-adapt
branch
3 times, most recently
from
d8c6078 to
b317d8c
Compare
yuanzhg078
reviewed
Jun 12, 2026
Fengli5355
force-pushed
the
asu-store-adapt
branch
from
b317d8c to
ae82dae
Compare
Infinite666
reviewed
Jun 12, 2026
Infinite666
reviewed
Jun 12, 2026
Infinite666
reviewed
Jun 12, 2026
Infinite666
reviewed
Jun 12, 2026
Infinite666
reviewed
Jun 12, 2026
[Fix] Remove unused throw-catch when configuring TransProviderType and add an unsupported type.
Fengli5355
force-pushed
the
asu-store-adapt
branch
from
8b0eeac to
f836182
Compare
ygwpz
reviewed
Jun 13, 2026
ygwpz
reviewed
Jun 13, 2026
ygwpz
reviewed
Jun 13, 2026
ygwpz
reviewed
Jun 13, 2026
ygwpz
reviewed
Jun 13, 2026
[Refactor] AsuStore makes key hash only by blockId now; [Refactor] Move PatchFakeBackendTransportConfig into asu_store.cpp:BuildTransportConfig where transProvider has type FAKE. Remove fake_backend.h/cpp; [Refactor] Make fake_trans_provider.cpp simplier when making FakeTransProviderConfig.
Fengli5355
force-pushed
the
asu-store-adapt
branch
from
f836182 to
5cab32c
Compare
Infinite666
approved these changes
Jun 15, 2026
ygwpz
reviewed
Jun 15, 2026
ygwpz
reviewed
Jun 15, 2026
mag1c-h
approved these changes
Jun 15, 2026
mag1c-h
merged commit 85c5094
into
ModelEngine-Group:feature_26h1
14 of 16 checks passed
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Purpose
Adapt AsuStore into Connector in order to run an e2e inference.
Modifications
Adaptations for AsuStore:
(1) Add a .yaml config file;
(2) Modify build/compile scripts;
(3) Define "shard_size" and "tensor_size" the same as "block_size" for AsuStore;
(4) Introduce fakebackend in AsuTransport and AsuStore;
Fixes of FakeBackend:
(1) FakeBackend Adds deviceId config;
(2) FakeBackend Adds aclrtInit and aclrtSetDevice. (Reset and Finalize are not added yet)
(3) Correct batchNum value in FakeBackend::CompleteDelete and CompleteExist;
(4) Correct direct NPU I/O in FakeBackend, and use aclrtMemcpy instead;
(5) Correct FakeBackend::Exist which lacks existing_key_number.
Add some error logs for AsuClient.
Change the way of hashing to try to avoid hash collision.
Adapt to BufferManager changes PR#1014.
Fix of AsuStore:
Correct AsuStore::Dump by adding WaitPrerequisiteEvent before submitting the task.
Further fixes:
(1) Rename config file as ucm_config_asu.yaml;
(2) AsuStore makes key hash only by blockId now;
(3) Move PatchFakeBackendTransportConfig into asu_store.cpp:BuildTransportConfig where transProvider has type FAKE.
(4) Remove fake_backend.h/cpp;
(5) Simplify fake_trans_provider.cpp when making FakeTransProviderConfig.
Test
examples/offline_inference.pywith asu store type shares the same output with cache|posix. (Basic correctness)"Xiamen University" Example test passes. (Single concurrency)
multifieldqa-zh.jsonl produces similar outputs in 2 runs in a row. (32 concurrency）