Rebase to Upstream#31

Closed

gmlwns2000 wants to merge 639 commits intodeepauto/devfrom

feat/rebase-0912

This pull request is big! We're only showing the most recent 250 commits

Commits on Sep 6, 2025

Commits on Sep 7, 2025

Commits on Sep 8, 2025

Commits on Sep 9, 2025

Commits on Sep 10, 2025

Commits on Sep 11, 2025

Commits on Sep 12, 2025

Support HiP Attention

committed
Improve for review comments

daniel-geon-park
authored and
gmlwns2000
committed
remove rope reuse code

daniel-geon-park
authored and
gmlwns2000
committed
update exaone

daniel-geon-park
authored and
gmlwns2000
committed
add error messages for unsupported models for hip attention

daniel-geon-park
authored and
gmlwns2000
committed
Implement abstract methods

daniel-geon-park
authored and
gmlwns2000
committed
fix eagle draft cuda graph runner

daniel-geon-park
authored and
gmlwns2000
committed
support deepseek-v2

daniel-geon-park
authored and
gmlwns2000
committed
giupdate pyproject.toml to include hip-attn

daniel-geon-park
authored and
gmlwns2000
committed
fix test flag

daniel-geon-park
authored and
gmlwns2000
committed
move for loop over batch dim into hip-attention library

daniel-geon-park
authored and
gmlwns2000
committed
feat: bump hip-attn

kbumsik
authored and
gmlwns2000
committed
bump hip-attn version

daniel-geon-park
authored and
gmlwns2000
committed
cleanup rebase

daniel-geon-park
authored and
gmlwns2000
committed
fix for MLA

daniel-geon-park
authored and
gmlwns2000
committed
fix args

daniel-geon-park
authored and
gmlwns2000
committed
update tests

daniel-geon-park
authored and
gmlwns2000
committed
bump hip-attn version

daniel-geon-park
authored and
gmlwns2000
committed
cleanup after rebase

daniel-geon-park
authored and
gmlwns2000
committed
remove redundant code

daniel-geon-park
authored and
gmlwns2000
committed
fix
gmlwns2000
committed
fix bug
gmlwns2000
committed
handling chunk attention in hip
gmlwns2000
committed
PASSKEY, you should remove before PR
gmlwns2000
committed
fix
gmlwns2000
committed
fix
gmlwns2000
committed
fix
gmlwns2000
committed
fix
gmlwns2000
committed
cleaning
gmlwns2000
committed
fmt
gmlwns2000
committed
support qwen2.5 vL
gmlwns2000
committed
cleanup

daniel-geon-park
authored and
gmlwns2000
committed
run auto format

daniel-geon-park
authored and
gmlwns2000
committed
fix
gmlwns2000
committed
bug fix
gmlwns2000
committed
support video input for qwen-vl

mickqian
authored and
gmlwns2000
committed
rebase

mickqian
authored and
gmlwns2000
committed
simplify processor

mickqian
authored and
gmlwns2000
committed
fix bug
gmlwns2000
committed
fix
gmlwns2000
committed
fix bug
gmlwns2000
committed
fix
gmlwns2000
committed
fix for qwen2 3 moe
gmlwns2000
committed
fix
gmlwns2000
committed
fix
gmlwns2000
committed
support chunked
gmlwns2000
committed
fix
gmlwns2000
committed
fix
gmlwns2000
committed
fix
gmlwns2000
committed
fix for qwen3
gmlwns2000
committed
fix
gmlwns2000
committed
fix
gmlwns2000
committed
fix
gmlwns2000
committed
fix
gmlwns2000
committed
fix

kbumsik
authored and
gmlwns2000
committed
fix

kbumsik
authored and
gmlwns2000
committed
fix

kbumsik
authored and
gmlwns2000
committed
fix

kbumsik
authored and
gmlwns2000
committed
fix

kbumsik
authored and
gmlwns2000
committed
fix

kbumsik
authored and
gmlwns2000
committed
fix

kbumsik
authored and
gmlwns2000
committed
dense decode ...?

kbumsik
authored and
gmlwns2000
committed
fix
gmlwns2000
committed
fix deepseekv2
gmlwns2000
committed
fix
gmlwns2000
committed
temporary fix
gmlwns2000
committed
fix bug
gmlwns2000
committed
fix
gmlwns2000
committed
fmt
gmlwns2000
committed
handle mla spec decode
gmlwns2000
committed
fmt
gmlwns2000
committed
fix bug
gmlwns2000
committed
fix
gmlwns2000
committed
fix radix
gmlwns2000
committed
fmt
gmlwns2000
committed
fix deepseek bug
gmlwns2000
committed
fmt
gmlwns2000
committed
support fp8
gmlwns2000
committed
fix
gmlwns2000
committed
fix
gmlwns2000
committed
improve warmup
gmlwns2000
committed
fix for fp8
gmlwns2000
committed
run
gmlwns2000
committed
add delta
gmlwns2000
committed
support fp8e4m3
gmlwns2000
committed
fa3 decode is handled in hip-attention lib
gmlwns2000
committed
update
gmlwns2000
committed
add self extend scale
gmlwns2000
committed
fmt
gmlwns2000
committed
this is bad idea...
gmlwns2000
committed
fix
gmlwns2000
committed
fix
gmlwns2000
committed
support GLM45
gmlwns2000
committed
fix warmup
gmlwns2000
committed
handling gpt-oss sk
gmlwns2000
committed
fmt
gmlwns2000
committed
support gpt oss
gmlwns2000
committed
fix gpt-oss
gmlwns2000
committed
fix
gmlwns2000
committed
add bug message
gmlwns2000
committed
fmt
gmlwns2000
committed
fmt
gmlwns2000
committed
fix
gmlwns2000
committed
fmt
gmlwns2000
committed
fix
gmlwns2000
committed
fmt
gmlwns2000
committed
fix
gmlwns2000
committed
add global var
gmlwns2000
committed
fix
gmlwns2000
committed
suport dual chunk attention
gmlwns2000
committed
fix
gmlwns2000
committed

Commits on Sep 25, 2025

try to support qwen3 next
gmlwns2000
committed