Support transformers 4.43 #1971
echarlaix
merged
32 commits into
main
from
support-transformers-4.43
fix bt bark test
9333b586
setup
4dda6df1
patch clip models for sd
5926bc56
infer ort model dtype property from inputs dtypes
c2a5c03a
patch all clip variants
b610212f
device setter
99230847
bigger model for now
0cb6be75
fix device attribution
88831a59
onnx opset for owlvit and owlv2
a1f838cc
model dtype
b8f5f323
revert
81d02271
use model part dtype instead
82a28799
no need for dtype with diffusion pipelines
d2a15b54
revert
c7610268
fix clip text model with projection not outputting hidden states
0eb5dce0
whisper generation
f568bf63
echarlaix
commented
on 2024-07-26
fix whisper, support cache_position, and using transformers whisper g…
92ea60b3
style
170eaba4
create cache position for merged decoder and fix test for non whisper…
991b66b8
typo
8f8e6ca3
Merge branch 'main' into support-transformers-4.43
e5934b38
conditioned cache position argument
96bdde14
update whisper min transformers version
9d09389c
compare whisper ort generation with transformers
056e4509
Merge branch 'support-transformers-4.43' of https://github.com/huggin…
b3d91811
fix generation length for speech to text model type
825cc6d4
echarlaix
commented
on 2024-07-30
cache position in whisper only with dynamic axis decoder_sequence_length
3fe0cac1
IlyasMoutawwakil
commented
on 2024-08-01
use minimal prepare_inputs_for_generation in ORTModelForSpeechSeq2Seq
b3948b93
remove version restrictions on whisper
2f69a8ad
comment
4cc1065b
fix
8077ded1
simpler
aa9b9d6f
Reviewers
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub