merge main to stable#60

Closed

peaceforeverCN wants to merge 53 commits intostablefrom

Commits on Sep 1, 2025

add vllm 0.10.0 support

linhu-nv
authored and
peaceforeverCN
committed

Commits on Sep 5, 2025

radix tree c++ impl (#70)

charliecgxu
authored and
peaceforeverCN
committed
sync kernel launch

Luis-xu
authored and
peaceforeverCN
committed
kvmanager refactor (#73)

authored and
peaceforeverCN
committed
feat: add support release wheel (#77)

lilgao
authored and
peaceforeverCN
committed
update unit tests for new version (#79)

zhuofan1123
authored and
peaceforeverCN
committed
enable profile in release build

charliecgxu
authored and
peaceforeverCN
committed
rename functions

zhuofan1123
authored and
peaceforeverCN
committed
add evict_ratio in cache config, default is 0
peaceforeverCN
committed
update benchmark worker (#82)

zhuofan1123
authored and
peaceforeverCN
committed
fix broken cpp radix tree support for cache engine (#84)

charliecgxu
authored and
peaceforeverCN
committed
ci: trigger on main and dev

lilgao
authored and
peaceforeverCN
committed
fix direct io

zhuofan1123
authored and
peaceforeverCN
committed
quickfix for return type of reduce_tensor

linhu-nv
authored and
peaceforeverCN
committed
fix bug

gz944367214
authored and
peaceforeverCN
committed
fix status bug

zhuofan1123
authored and
peaceforeverCN
committed
Using ring buffer in transfer engine to manage the src and dst block ids instead of using pin memory function inner the launch kernel for reducing the bubble

Luis-xu
authored and
peaceforeverCN
committed
refine ring_buffer and apply it to all workers

Luis-xu
authored and
peaceforeverCN
committed
rename PinnedMemoryRing to SharedMemoryRing

Luis-xu
authored and
peaceforeverCN
committed
allow to exceed the max_block_num

zhuofan1123
authored and
peaceforeverCN
committed
refactor: use hash to allocate buffer && no wait for free slot

zhuofan1123
authored and
peaceforeverCN
committed
allow different tp ranks have different num_gpu_blocks

linhu-nv
authored and
peaceforeverCN
committed
fix

linhu-nv
authored and
peaceforeverCN
committed
create arrays of gpu_block infos in c++ to avoid invalid ptrs

linhu-nv
authored and
peaceforeverCN
committed
vllm v0.10.1.1 adapter

gz944367214
authored and
peaceforeverCN
committed
fix bug

zhuofan1123
authored and
peaceforeverCN
committed
server-client mode works now (#92)

linhu-nv
authored and
peaceforeverCN
committed
[docs] change vllm adapter README
peaceforeverCN
committed
[docs] add stable branch introduce
peaceforeverCN
committed
[doc] add CONTRRIBUTING.md
peaceforeverCN
committed
Merge pull request #3 from taco-project/docs/move_vllm_patch
linhu-nv
authored

Commits on Sep 9, 2025