[0.9.1][Feature]Moe alltoallv communication optimization for unquantized RL training sence & alltoallv support dpo#1547

Merged

ganyi1996ppo merged 68 commits intovllm-project:v0.9.1-devfrom

weijinqian0:v0.9.1-dev

Jul 14, 2025

Commits on Jul 1, 2025

[Feature]Moe alltoallv communication optimization for unquantized RL training sence & alltoallv support dpo
weijinqian_v1
committed
[Feature]Moe alltoallv communication optimization for unquantized RL training sence & alltoallv support dpo
weijinqian_v1
committed
[Feature]Moe alltoallv communication optimization for unquantized RL training sence & alltoallv support dpo
weijinqian_v1
committed
[Feature]Moe alltoallv communication optimization for unquantized RL training sence & alltoallv support dpo
weijinqian_v1
committed
[Feature]Moe alltoallv communication optimization for unquantized RL training sence & alltoallv support dpo
weijinqian_v1
committed
[Feature]Moe alltoallv communication optimization for unquantized RL training sence & alltoallv support dpo
weijinqian_v1
committed
[Feature]Moe alltoallv communication optimization for unquantized RL training sence & alltoallv support dpo
weijinqian_v1
committed
[Feature]Moe alltoallv communication optimization for unquantized RL training sence & alltoallv support dpo
weijinqian_v1
committed
[Feature]Moe alltoallv communication optimization for unquantized RL training sence & alltoallv support dpo
weijinqian_v1
committed
[Feature]Moe alltoallv communication optimization for unquantized RL training sence & alltoallv support dpo
weijinqian_v1
committed
[Feature]Moe alltoallv communication optimization for unquantized RL training sence & alltoallv support dpo
weijinqian_v1
committed
[Feature]Moe alltoallv communication optimization for unquantized RL training sence & alltoallv support dpo
weijinqian_v1
committed
[Feature]Moe alltoallv communication optimization for unquantized RL training sence & alltoallv support dpo
weijinqian_v1
committed
[Feature]Moe alltoallv communication optimization for unquantized RL training sence & alltoallv support dpo
weijinqian_v1
committed

Commits on Jul 2, 2025

[Feature]Moe alltoallv communication optimization for unquantized RL training sence & alltoallv support dpo
weijinqian_v1
committed

Commits on Jul 8, 2025

handle conflict
weijinqian_v1
committed

Commits on Jul 9, 2025

add st:qwen3
weijinqian_v1
committed
add st for moe token dispatcher
weijinqian_v1
committed
fix bug

harygo22
authored and
weijinqian_v1
committed
add st for moe token dispatcher
weijinqian_v1
committed
add moe_block: AscendSparseMoeBlock
weijinqian_v1
committed
add moe_block: AscendSparseMoeBlock
weijinqian_v1
committed
add moe_block: AscendSparseMoeBlock
weijinqian_v1
committed
[0.9.1][Perf] Optimize the number of rope-related index selections in deepseek. (#1614)

whx-sjtu
authored and
weijinqian_v1
committed
[BUGFIX] FIX mtp accuraccy when temperture is not 0 (#1632)

JC-ut0
authored and
weijinqian_v1
committed
add mc2 mask (#1642)

weiguihua2
authored and
weijinqian_v1
committed
[cherry-pick] static EPLB fix bug, add unit test to v0.9.1-dev (#1667)

songshanhu07
authored and
weijinqian_v1
committed
revert

harygo22
authored and
weijinqian_v1
committed
fix bug

harygo22
authored and
weijinqian_v1
committed
fix a bug

harygo22
authored and
weijinqian_v1
committed
fix a bug

harygo22
authored and
weijinqian_v1
committed
ut test

harygo22
authored and
weijinqian_v1
committed
liscens & fix dsk dbo.

harygo22
authored and
weijinqian_v1
committed
handle conflict
weijinqian_v1
committed
handle code clean
weijinqian_v1
committed
handle code clean
weijinqian_v1
committed
handle code clean
weijinqian_v1
committed
handle code clean
weijinqian_v1
committed

Commits on Jul 10, 2025

Merge branch 'v0.9.1-dev' into v0.9.1-dev
harygo22
authored
fix comment
harygo22
committed
handle code clean
weijinqian_v1
committed
handle code conflict
weijinqian_v1
committed
fix init
harygo22
committed
remove files & move sparsemoeblock to ops
harygo22
committed

Commits on Jul 11, 2025

remove test
harygo22
committed
clean code
harygo22
committed
fix clean code
harygo22
committed
typo
harygo22
committed
renaming cuda sync point
harygo22
committed
Merge branch 'v0.9.1-dev' of https://github.com/weijinqian0/vllm-ascend into v0.9.1-dev
weijinqian_v1
committed
handle code clean
weijinqian_v1
committed
handle code clean
weijinqian_v1
committed
handle code clean
weijinqian_v1
committed

Commits on Jul 12, 2025

handle clean code
weijinqian
committed
handle clean code
weijinqian
committed
handle clean code
weijinqian
committed
handle clean code
weijinqian
committed
handle clean code
weijinqian
committed
handle clean code
weijinqian
committed
handle clean code
weijinqian
committed
handle clean code
weijinqian
committed
handle clean code
weijinqian
committed
handle clean code
weijinqian
committed
handle clean code
weijinqian
committed
handle code clean
weijinqian_v1
committed
handle code conflict
weijinqian_v1
committed
handle code clean
weijinqian_v1
committed

Commits on Jul 14, 2025

fix header
harygo22
committed