Pooling methods for cross-encoder
Nobody has claimed this yet.
Assessment
- Difficulty
- 5/5
- Estimated time
- Over a week
- Newbie friendliness
- 30/100
- Issue type
- Feature
- Clarity
- Needs clarification
- Activity status
- Stale
- Tech stack
- python
- Domain
- machine-learning
Research direction
Start by reviewing the CrossEncoder.predict() and SentenceTransformer usage, then trace the _classifier_hook path and the train_classifier flag. Compare the current embedding from token 0 with the pooling methods that need to be supported; done should define and implement the intended pooling behavior without relying on an assumed CLS token.
Written by the indexing model from the issue text.
Description
Я говорю про то чтобы не использовать _classifier_hook можно просто сделать SentenceTranformer, а не CrossEncoder
причины почему мы используем хук:
* это работает :) * нам нужно использовать один трансформер в двух режимах в зависимости от флага `train_classifier`: с претренированной головой и со своей кастомнойПретренированную голову точно можно ипользовать с помощью
CrossEncoder.predict(), а о том что ее можно использовать и с помощью какого-то метода классаSentenceTransformerмне неизвестно. Если это и можно сделать, то наверное это что-то экзотическое и не задокументировано в самой библиотекеsentence_transformers
Да, нельзя использовать доп голову. Просто сейчас у нас реализован получение эмбеда из 0 токена (предположим, что это CLS) но не все модели обучались с ним и не факт, что это хорошо работать будет
Originally posted by @Samoed in https://github.com/deeppavlov/AutoIntent/issues/88#issuecomment-2585304933
- Dominant language
- Python
- Stars
- 53
- Forks
- 16
- PR merge metrics
- No merged PRs in 30d
Getting set up
- No Dockerfile or Docker Compose file
- No pull request template
- Read the contributing guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
More from deeppavlov/AutoIntent
-
`OptimizationConfig.seed` is `PositiveInt` — `seed=0` is rejected while `Pipeline(seed=0)` accepts itPossibly taken @Gambit-Checkmate claimed this 21 days ago. Openbug good first issue
Difficulty 2/5 1-3 hours Newbie friendliness 85/100
deeppavlov/AutoIntent#352 ·
-
Extract `BaseAPIDescriptionScorer` shared by `LLMDescriptionScorer` and `TypeSafeDescriptionScorer`Openenhancement
Difficulty 4/5 3-5 days Newbie friendliness 48/100
deeppavlov/AutoIntent#357 ·
-
bug
Difficulty 4/5 3-5 days Newbie friendliness 52/100
deeppavlov/AutoIntent#356 ·
-
enhancement
Difficulty 3/5 1-2 days Newbie friendliness 72/100
deeppavlov/AutoIntent#355 · 1 comment ·
-
`LLMDescriptionScorer` runs cache hits through the `max_per_second` limiter — 10 s per 100 cached utterancesPossibly taken @kayaal34 claimed this 21 days ago. Openenhancement
Difficulty 4/5 3-5 days Newbie friendliness 58/100
deeppavlov/AutoIntent#354 ·
All issues in deeppavlov/AutoIntent
Similar issues
-
bug
Difficulty 2/5 1-3 hours Newbie friendliness 78/100
topoteretes/cognee#5647 · 1 comment ·
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 62/100
Sendspin/sendspin-python-cli#291 ·
Maintainers usually reply within 6 days
-
Assertion-shape guard fails on development: vacuous recorded-iteration assertion in the assetLinks batch_get helper testPossibly taken A pull request linked to this issue is open or already merged. Openbug
Difficulty 2/5 1-3 hours Newbie friendliness 72/100
awslabs/visual-asset-management-system#414 ·
Maintainers usually reply within 1 day
-
bug v1 v2
Difficulty 2/5 1-3 hours Newbie friendliness 75/100
modelcontextprotocol/python-sdk#3670 · 1 comment ·
Maintainers usually reply within 1 day
-
Difficulty 2/5 1-3 hours Newbie friendliness 82/100
aicell-lab/bioengine#232 ·
Maintainers usually reply within 1 day