[Cache] Native mamba & hybrid cache #44950
Cyrilvallez
wants to merge
55 commits into
main
from
clean-mamba-cache
add Cache and test on Mamba
db8e4ffc
fix
9d525982
fix
659beee9
fix
29b91abf
fix
3e02650b
fix
fb883452
final fix
1aeddfae
test hybrid with jamba
35db152b
fix tests
a50293ca
fixes
1607fe29
fix
ddc198af
fix
bae4a78d
fix
984b578a
combine both types + zambas
cac5d17c
add config mapèping
bd8f9e9a
adjust tests
b2f1bb8e
fix
77958087
fix
18685c60
fix
fcec6bc0
more models
b1df43f6
final mambas
fdb15791
config
b156adeb
finalize almost everything
330e397d
simplify tests
b60c6f59
simplify tests further
0e8ca285
fix tests
c2ddcf9b
oupsi
b23708fa
fix
18feef28
fix broken no_split_modules
ce92f3dc
fix
ab4472bf
fixes
08e62658
fix
66d07164
fix
c86f9bb8
fixes
ba1b7d68
add layer type
17856217
oupsi
f684133e
fix
8ca92a96
style
0d991d74
fix
670d09a3
Merge branch 'main' into clean-mamba-cache
bc99c9ad
fixes
63e0b939
final fix
eb018e7e
forgot those qwens
f8a07029
tests
fc27c37f
offloading
9c616dd7
much better static shape native design
6f85f549
oupsi
f4fc8017
adjustments in generate
6aca24ed
allow cudagraphs
13781f17
small oupsi
3df0d85a
Merge branch 'main' into clean-mamba-cache
39dae285
start renaming
eadcfa47
revert unrelated what are they doing here
908f0da1
Merge branch 'main' into clean-mamba-cache
cf870665
more renaming
86de2bc8
Login to write a write a comment.
Login via GitHub