Joao Gante

Generate: support passing position_ids

2

Thank you @tengomucho, for uncovering this bug. ### The problem In a nutshell, passing the correct `position_ids` to `generate` should result in exactly the same results as not passing them....

DO NOT MERGE: generate compatible with torch.compile(fullgraph=True)

1

# What does this PR do? This PR is a 🔪 mangled🔪 version of `generate` where `torch.compile(model.generate, fullgraph=True)` works and returns the same values. It should NOT be merged, but...

Whisper: fix asr pipeline with seq2seq assistant model

1

# What does this PR do? Fixes #29869 Fixes #30407 Fixes #30611 (related PR: #30637) The ASR pipeline was preparing the encoder outputs before `generate` (and not passing `input_features`), but...

SlidingWindowCache: reduce differences to other Cache classes

1

# What does this PR do? Follow up to #30642: this PR aims at reducing the differences between `SlidingWindowCache` and `StaticCache`, such that long-term maintenance becomes easier. Fewer attributes/functions =...

dependencies: `keras-nlp<0.14` pin

9

# What does this PR do? On the latest `keras-nlp` [release](https://github.com/keras-team/keras-nlp/releases/tag/v0.14.0) -- "KerasNLP no longer supports Keras 2.". This means we don't support this version, as we only support Keras...

Gemma 2: toggle `_supports_cache_class`

1

# What does this PR do? It should be `True`

Docker image: keras-nlp<0.14 pin

3

Temporary PR to update the docker image