PR #38579 blt wip - SemanticDiff

itazap force pushed from f9d97099 to a96f92f9 224 days ago

itazap force pushed from a3222d96 to 0a16089c 217 days ago

LysandreJik commented on 2025-06-23

LysandreJik commented on 2025-06-25

LysandreJik commented on 2025-06-27

itazap force pushed from 76f1ae81 to 3772dc76 196 days ago

itazap force pushed from d8ea1780 to 367878ae 191 days ago

LysandreJik commented on 2025-07-10

itazap force pushed from 6e4d25f6 to a08cf708 184 days ago

LysandreJik marked this pull request as ready for review 183 days ago

itazap force pushed from 8343f98e to 84b991a6 183 days ago

itazap force pushed from a7606314 to 3a747131 182 days ago

itazap force pushed from f6164ab6 to 4f2348bd 182 days ago

Cyrilvallez commented on 2025-07-18

itazap force pushed from 6cfe9cf9 to 4fbbe826 178 days ago

itazap force pushed from 38b616ea to 6e524227 176 days ago

itazap force pushed from dba4d6f0 to 0715701b 171 days ago

itazap force pushed from 0715701b to 2b318733 171 days ago

itazap force pushed from 3baf0b00 to d2f5830d 171 days ago

itazap requested a review from

Cyrilvallez 170 days ago

itazap force pushed from 84683d56 to 563d5e42 169 days ago

Cyrilvallez commented on 2025-08-05

itazap requested a review from

Cyrilvallez 160 days ago

Cyrilvallez commented on 2025-08-13

itazap commented on 2025-08-18

itazap requested a review from

Cyrilvallez 149 days ago

itazap force pushed from 7b569493 to 02be2848 149 days ago

Cyrilvallez commented on 2025-08-20

Cyrilvallez commented on 2025-08-21

itazap force pushed from 450a00f5 to 6a5f7674 146 days ago

itazap requested a review from

Cyrilvallez 146 days ago

Cyrilvallez approved these changes on 2025-08-28

ArthurZucker commented on 2025-08-28

itazap requested a review from

Cyrilvallez 135 days ago

itazap requested a review from

ArthurZucker 126 days ago

ArthurZucker approved these changes on 2025-09-12

itazap requested a review from

ArthurZucker 122 days ago

ArthurZucker approved these changes on 2025-09-18

blt wip

08261523

cpu version

62019471

cpu friendly with full entropy model (real time patching)

58c4a4e7

adding config file instead of args file

1d00859a

enable MPS

bdb6ceef

refactoring unused code

131f9608

single config class in config file

fb1d11ba

inherit from PreTrainedModel

1eab6a4a

refactor LMTransformer --> BLTPatcher

bc2aeb74

add conversion script

907eca15

load from new checkpoing with form_pretrained

c4b17753

fixed demo from_pretrained

fececd1c

clean up

f2604f30

clean a few comments

12c000e8

cleanup folder

4b2185db

clean up dir

ad8c7a89

cleaned up modeling further

aff63d6b

rename classes

f552e275

adding transformers Attention class and RotaryEmbedding class

2b9dd64f

exchanged blt modules for transformers modules: attention, rotary_emb…

f25a99b5

seperate out patcher config, update modeling and conversion script

73f7e169

rename vars to be more transformers-like

8d4df991

rm unused functions

d938a2f1

adding cross attention from transformers

3bcfc03d

pass arg

2a7778c3

rename weights

9ed04fda

updated conversion script

e6c7b68d

overwritten commit! fixing PR

e3fdebba

apply feedback

438e2e26

adding BLTRMSNorm like Llama

8ecda842

add repeat_kv and eager_attention_forward copied from

ceb3d8e2

BLTMLP identical to MllamTextMLP

2102d325

clean up some args'

50d25036

more like mllama, but busier inits

66bcddb9

BLTTransformerLayer config

5bcc11d9

decoder, encoder, global configs

aa03d78f

wip working on modular file

494b4881

cleaning up patch and configs

477406e9

clean up patcher helpers

13a79a52

clean up patcher helpers further

f686d0b3

missed file

a260bb1b

added tied weights keys

0b9db70f

BLTForCausalLM

107e26d5

adding files after add-new-model-like

d9d6d730

update demo

3e8dc1e5

working on tests

e6bc6398

first running integration tests

7c352ae7

added integration tests

c32f692a

adding tokenization tests, integration tests, and cleaned up tokeniza…

e7238303

tokenizer clean up

3a22e052

modular file

d7b57214

fixing rebase

359ecf11

ruff

4e528374

adding correct basemodel output and updating config with checkpoint v…

fed99588

BLTModelTests git status

3ab48a66

enabling inputs_embeds, although won't be equal to input_ids since ne…

6f931992

fix sdpa == causal tests

07816a2a

fix small model test and some gradient checkpointing

db4cc785

skip training GC tests

c20e4849

fix test

d2dab129

updated modular

2dce4bf5

update modular

e64ed907

ruff

77380053

adding modular + modeling

8b2a2383

modular

0c23353f

more modern is_casual check

141d788b

cleaning up modular

fd1dd4ab

more modular reduction

82bff4e9

ruff

6f174745

modular fix

c191396e

fix styling

50c0353b

return 2

36a95538

return 2

132830e0

fix some tests

9f3a3b42

fix bltcrossattention after modular break

80953034

some fixes / feedback

562c03ad

try cache generate fix

fc1e7bfa

try cache generate fix

8add2447

fix generate tests

f198de79

attn_impl workaround

ab4d2cae

refactoring to use recent TransformersKwargs changes

a00ce1de

fix hidden_states shape test

1df0b6a2

refactor to new outputs

3f7d5cdd

simplify outputs a bit

22a511a7

rm unneeded decoderlayer overwriting

0239f773

rename blt

926fb092

forgot tokenizer test renamed

232d245f

Reorder

703fab75

Reorder

ec9b4c08

working on modular

3117a038

updates from modular

eb4cd414

new modular

c9e30fd9

ruff and such

3b2e3e85

update pretrainedmodel modular

2ded41e0

using cohere2 apply_rotary_pos_emb

cd7d1a8d

small changes

01835380

apply feedback r2

cb91d0e9

fix cross_attention

f51e2f47

apply more feedback

22a20f29

update modeling fix

39be4145

load submodules from pretrainedmodel

6ecc6ff6

set initializer_range to subconfigs

eea290d4

rm cross_attnetion_states pass when not needed

294b80dc

add 7b projection layer support

9ec7b28f

check repo

2f9ab611

make copies

3e280825

lost cohere2 rotate_half

52fa9871

ruff

f25630ca

copies?

26706e59

don't tie weights for submodules

35dde6ef

tie weights setting

f855e52f

check docstrings

966e2f03

apply feedback

5513a6a6

rebase

29144c79

rebased modeling

8869cc19

update docs

f3e62f00

applying feedback

cab52b59

few more fixes

d45f2603

fix can_record_outputs

7ccff57e

fast tokenizer

90a9a2fb

no more modulelist

180042d9

tok auto

c495819b

rm tokenizersss

5607b5ad

fix docs

8085a95a

ruff

4272552d

itazap force pushed from 616293d8 to 4272552d 119 days ago

fix after rebase

05a5b496

fix test, configs are not subscriptable

d983e72b

Merge branch 'main' into blt_wip

17f91b97

ArthurZucker merged ddfa3d44 into main 118 days ago

ArthurZucker deleted the blt_wip branch 118 days ago

transformers
blt wip
#38579

Merged

blt wip #38579

transformers blt wip #38579 Merged

blt wip #38579

transformers
blt wip
#38579

Merged