transformers
Remove ambiguous `padding_mask` and instead use a 2D->4D Attn Mask Mapper
#26792

Merged

Remove ambiguous `padding_mask` and instead use a 2D->4D Attn Mask Mapper #26792

patrickvonplaten merged 28 commits into main from attn_mask_converter

[Attn Mask Converter] refactor attn mask

3b5dfb75

50326280

fxmarty commented on 2023-10-13

patrickvonplaten changed the title ~~[Attn Mask Converter] refactor attn mask~~ [WIP][Attn Mask Converter] refactor attn mask 2 years ago

younesbelkada commented on 2023-10-13

ArthurZucker commented on 2023-10-16

Apply suggestions from code review

0bfbc1ad

improve

d173ce30

improve

1d230de2

rename

e542a6d7

better cache

0256cd5e

patrickvonplaten changed the title ~~[WIP][Attn Mask Converter] refactor attn mask~~ Remove ambiguous `padding_mask` and instead use an AttentionMaskConverter class 2 years ago

patrickvonplaten changed the title ~~Remove ambiguous `padding_mask` and instead use an AttentionMaskConverter class~~ Remove ambiguous `padding_mask` and instead use a 2D->4D Attn Mask Mapper with Cache 2 years ago

renaming

7e50ec0e

improve more

30fc4c36

improve

2301d6ba

fix bug

4cfb7cb4

patrickvonplaten commented on 2023-10-16

finalize

bdb39ae3

patrickvonplaten changed the title ~~Remove ambiguous `padding_mask` and instead use a 2D->4D Attn Mask Mapper with Cache~~ Remove ambiguous `padding_mask` and instead use a 2D->4D Attn Mask Mapper / Cache 2 years ago

patrickvonplaten requested a review from

LysandreJik 2 years ago

patrickvonplaten requested a review from

fxmarty 2 years ago

patrickvonplaten requested a review from

ArthurZucker 2 years ago

patrickvonplaten requested a review from

younesbelkada 2 years ago

fxmarty commented on 2023-10-16

ArthurZucker approved these changes on 2023-10-16

LysandreJik commented on 2023-10-18

younesbelkada commented on 2023-10-18

Merge branch 'main' of https://github.com/huggingface/transformers in…

67bd54e6

Merge branch 'attn_mask_converter' of https://github.com/huggingface/…

2c42f7df

make style & make fix-copies

4387ab8d

correct more

068ed57e

start moving attention_mask

d18268a8

fix llama

4a99e436

improve falcon

2fe66a0a

d03d8a14

improve more

9e553ac7

improve more

431b3a83

patrickvonplaten commented on 2023-10-19

Update src/transformers/models/owlv2/modeling_owlv2.py

f8b2e4e7

make style

0338ffea

Merge branch 'attn_mask_converter' of https://github.com/huggingface/…

eb953152

patrickvonplaten commented on 2023-10-19

make style

ae3eb2e7

patrickvonplaten requested a review from

fxmarty 2 years ago

patrickvonplaten requested a review from

LysandreJik 2 years ago

patrickvonplaten requested a review from

younesbelkada 2 years ago

patrickvonplaten requested a review from

ArthurZucker 2 years ago

patrickvonplaten commented on 2023-10-19

rename to converter

b438dc84

patrickvonplaten changed the title ~~Remove ambiguous `padding_mask` and instead use a 2D->4D Attn Mask Mapper / Cache~~ Remove ambiguous `padding_mask` and instead use a 2D->4D Attn Mask Mapper 2 years ago

patrickvonplaten commented on 2023-10-19

Apply suggestions from code review

53399888

fxmarty commented on 2023-10-20

younesbelkada approved these changes on 2023-10-23

ArthurZucker commented on 2023-10-20

patrickvonplaten merged 33f98cfd into main 2 years ago

patrickvonplaten deleted the attn_mask_converter branch 2 years ago

Reviewers

ArthurZucker

younesbelkada

LysandreJik

fxmarty

Assignees

No one assigned

Labels

None yet

Milestone

No milestone

transformers Remove ambiguous `padding_mask` and instead use a 2D->4D Attn Mask Mapper #26792 Merged

Remove ambiguous `padding_mask` and instead use a 2D->4D Attn Mask Mapper #26792

transformers
Remove ambiguous `padding_mask` and instead use a 2D->4D Attn Mask Mapper
#26792

Merged