transformers
🚨 Always return Cache objects in modelings (to align with generate)
#39765

Merged

🚨 Always return Cache objects in modelings (to align with generate) #39765

manueldeprada merged 32 commits into huggingface:main from manueldeprada:no-from_legacy-init

watch the world burn

fe5f374d

fix models, pipelines

6567a14d

manueldeprada force pushed from 595a1a4f to 6567a14d 131 days ago

fix models that are both enc-dec and StandaloneDecoder

605ba142

gante dismissed these changes on 2025-08-05

manueldeprada marked this pull request as ready for review 131 days ago

Merge branch 'main' into no-from_legacy-init

0f4a7496

manueldeprada changed the title ~~[draft] No more using `from_legacy_cache` as initialization~~ Stop using `from_legacy_cache` as Cache initialization 131 days ago

Merge branch 'main' into no-from_legacy-init

85812b90

replace routing_weights.device.index with a

7cdd9487

Merge branch 'fix-cpu-tests' into no-from_legacy-init

375146fb

manueldeprada requested a review from

Cyrilvallez 131 days ago

Merge branch 'main' into no-from_legacy-init

c6fc969c

Cyrilvallez commented on 2025-08-12

cyril review

0d8d68be

Merge branch 'main' of github.com:huggingface/transformers into no-fr…

053cdf1c

raise error if None is passed

f725e218

fix xlm model

495ef608

Cyrilvallez dismissed these changes on 2025-08-12

fixes

00a1cd4a

ops

18dde861

more models

3bfcfae6

Merge branch 'main' of github.com:huggingface/transformers into no-fr…

d6fb95eb

fix flaubert

94e47120

fix bert test, dont use legacy

c7f336af

Merge branch 'main' into no-from_legacy-init

e051637f

fix pipelines definetely, sanitize cache's select_index, create tests

959271f3

manueldeprada changed the title ~~Stop using `from_legacy_cache` as Cache initialization~~ Stop using `from_legacy_cache` as Cache init, make `pipeline` use Caches, fix and test `Cache.select_indices` 122 days ago

fix tests/trainer/test_trainer.py::TrainerIntegrationTest::test_evalu…

0ad4c8f8

Revert "fix tests/trainer/test_trainer.py::TrainerIntegrationTest::te…

24949fac

Cyrilvallez commented on 2025-08-14

Cyrilvallez dismissed their stale review 122 days ago

unrelated changes need clarification

Cyrilvallez dismissed their stale review 122 days ago

unrelated changes need clarification

Cyrilvallez changed the title ~~Stop using `from_legacy_cache` as Cache init, make `pipeline` use Caches, fix and test `Cache.select_indices`~~ 🚨 Return Cache objects in models when past_key_values are not provided (to align with generate) 122 days ago

skip test

89dbbe07

revert cache changes

04021bcf

ops

56a1b09f

manueldeprada commented on 2025-08-15

fix gpt2 test

35dd4dba

manueldeprada commented on 2025-08-15

make the error a warning

259de5a7

manueldeprada commented on 2025-08-15

manueldeprada requested a review from

Cyrilvallez 121 days ago

Cyrilvallez commented on 2025-08-15

remove kwargs and return_legacy_cache

4693a109

cont

260fab91

cyril review

9baf489a

Merge branch 'main' of github.com:huggingface/transformers into no-fr…

a0eaa4a4

fix reformer

28cd663f

manueldeprada requested a review from

Cyrilvallez 118 days ago

manueldeprada removed review request from

Cyrilvallez 118 days ago

manueldeprada requested a review from

gante 118 days ago

manueldeprada requested a review from

Cyrilvallez 118 days ago

Cyrilvallez changed the title ~~🚨 Return Cache objects in models when past_key_values are not provided (to align with generate)~~ 🚨 Always return Cache objects in modelings (to align with generate) 118 days ago

Cyrilvallez approved these changes on 2025-08-18

manueldeprada merged a36d51e8 into main 118 days ago

gante approved these changes on 2025-08-18

Reviewers

Cyrilvallez

gante

Assignees

No one assigned

Labels

None yet

Milestone

No milestone

transformers 🚨 Always return Cache objects in modelings (to align with generate) #39765 Merged

🚨 Always return Cache objects in modelings (to align with generate) #39765

transformers
🚨 Always return Cache objects in modelings (to align with generate)
#39765

Merged