huggingface/transformers

Pull Requests Commits

update cache shape

MekkCyber committed 139 days ago

fbd288fb

MekkCyber committed 139 days ago

eb714c4b

chaing v2 default

MekkCyber committed 139 days ago

028e6d6e

MekkCyber committed 139 days ago

b6322afd

ArthurZucker committed 139 days ago

04828074

comment latest sync for now

MekkCyber committed 139 days ago

3dc01c46

MekkCyber committed 139 days ago

d0be498d

MekkCyber committed 139 days ago

5e9b251a

MekkCyber committed 139 days ago

255bc3c9

MekkCyber committed 139 days ago

1fd77c3d

MekkCyber committed 139 days ago

5189c26e

MekkCyber committed 139 days ago

2721261e

kernel either works or not used

MekkCyber committed 139 days ago

571af3e2

figuring out why sdpa without kernel is broken

MekkCyber committed 139 days ago

51e115ac

remove mps device

MekkCyber committed 139 days ago

962f7c16

update flash_paged

ArthurZucker committed 139 days ago

fa3e0db0

I need to test on cuda, this shouldd work

ArthurZucker committed 139 days ago

3cbe8254

the shape expected by the kernel is not the sliced cache I reckon

ArthurZucker committed 139 days ago

6fbe4177

make sure we can compare without reshape kernel

ArthurZucker committed 139 days ago

9f47da14

ArthurZucker committed 139 days ago

da935a68

bunch of cleanup and let's not sample, easier to teset

ArthurZucker committed 139 days ago

f73db2b6

smaller for faster

ArthurZucker committed 139 days ago

c5765160

works with this :wink:

ArthurZucker committed 139 days ago

6eec32cf

ArthurZucker committed 139 days ago

cbe3e632

comment out stuff for now!

ArthurZucker committed 139 days ago

c2cf536c

losing my mind, test on cluster

MekkCyber committed 139 days ago

4d86a5bb

MekkCyber committed 139 days ago

ccf8a46c

MekkCyber committed 139 days ago

f205a9c9

fisrt integration

MekkCyber committed 139 days ago

458b3522

Update src/transformers/generation/continuous_batching.py

ArthurZucker committed 139 days ago

012d8f47

Older