[rebase]Dev-ucm-v1 rebase to develop#453

Merged

ygwpz merged 29 commits intoModelEngine-Group:developfrom

Lijiachen1018:dev-ucm-v1

Dec 2, 2025

Commits on Dec 1, 2025

[opt] refactor uc connector (#364)

ygwpz
authored and
lijiachen
committed
[Feat] Implement kv cache broadcast in MLA (#367)

harrisonyhq
authored and
lijiachen
committed
[feature] add ucm mock connector (#375)

ygwpz
authored and
lijiachen
committed
[Feat] Support get launch config from yaml (#377)

harrisonyhq
authored and
lijiachen
committed
[fix] refuse monkey patch (#383)

ygwpz
authored and
lijiachen
committed
[bugfix] fix gqa bug (#384)

ygwpz
authored and
lijiachen
committed
[bugfix] end == 0 bug (#385)

ygwpz
authored and
lijiachen
committed
[feature] optimize generate_tensor (#396)

ygwpz
authored and
lijiachen
committed
[Fix] fix mla bug when no broadcast in wait for save (#398)

harrisonyhq
authored and
lijiachen
committed
[feat]adapt GQA & modify config.yaml (#407)

qyh111
authored and
lijiachen
committed
[feat]Adapt vllm_ascend_0110 and Add configurable options (#415)

qyh111
authored and
lijiachen
committed
[patch]seprate sparse patch (#417)

Lijiachen1018
and
lijiachen19
committed
[bugfix]Support tensor parallelism across servers (#420)

qyh111
authored and
lijiachen
committed
[Feat] UCM supports metrics display online via Grafana and Promethues (#414)

authored and
lijiachen
committed
[feat]Merge develop to dev-ucm-v1 and fix code style (#428)

committed
add env variable ENABLE_SPARSE (#430)

Lijiachen1018
and
lijiachen19
committed
Fix(patch): fix patch for vllm-ascend (#433)

Lijiachen1018
and
lijiachen19
committed
[bugfix] fix accuracy problem when chunked prefill (#438)

qyh111
authored and
lijiachen
committed
[bugfix]fix num_schedule-tokens=1 (#442)

qyh111
authored and
lijiachen
committed
[fix]: Fix sparse patch (#444)

Lijiachen1018
and
lijiachen
committed
[bugfix] The Metrics module uses a non-existent variable self.rank (#445)

sumingZero
authored and
lijiachen
committed
[Feature]Add an access bandwidth test script for ucm_connector (#418)

zhou-haitao
authored and
lijiachen
committed
[bugfix]adapt vllm0.9.1 (#446)

qyh111
authored and
lijiachen
committed
[Fix]Set the multiprocessing start method of the test tool to 'spawn'. (#447)

zhou-haitao
authored and
lijiachen
committed
[fix] Adapt all sparse-attention methods to the new connector. (#441)

wangwenxin0312
authored and
lijiachen
committed
[docs] renew docs for v1 (#448)

Lijiachen1018
and
lijiachen19
committed
set version to 0.1.0 (#450)

qyh111
authored and
lijiachen
committed
[Feature] GSA adapt nfsStore (#451)

zbb200819
authored and
lijiachen
committed

Commits on Dec 2, 2025

fix codestyle
mag1c-h
committed