v0.7.0

[mattdangerw]

This release integrates KerasNLP and Kaggle Models. KerasNLP models will now work in Kaggle offline notebooks and all assets will quickly attach to a notebook rather than needing a slow download.

Summary

KerasNLP pre-trained models are now all made available through Kaggle Models. You can see all models currently available in both KerasCV and KerasNLP here. Individual model pages will include example usage and a file browser to examine all available assets for a model preset.

This change will not affect the existing usage of from_preset(). Statement like keras_nlp.models.BertClassifier.from_preset("bert_base_en") will continue to work and download checkpoints from the Kaggle Models hub.

An note on model saving—for saving support across Keras 2 and Keras 3, we recommend using the new Keras saved model format. You can use model.save('path/to/location.keras') for a full model and model.save_weights('path/to/location.weights.h5') for checkpoints. See the Keras saving guide for more details.

What's Changed

• Don't export model internals publicly by @mattdangerw in Don't export model internals publicly #1255
• Bump master branch version number to 0.7.0.dev0 by @mattdangerw in Bump master branch version number to 0.7.0.dev0 #1254
• Fix/allow different encoder and decoder feature dimensions in transformer decoder layer by @ferraric in Fix/allow different encoder and decoder feature dimensions in transformer decoder layer #1260
• Doc updates to switch branding to Keras 3 by @mattdangerw in Doc updates to switch branding to Keras 3 #1259
• Remove unused TPU testing for backbones by @mattdangerw in Remove unused TPU testing for backbones #1266
• Make gelu a function, not a lambda so it can be loaded without safe_mode=False by @calvingiles in Make gelu a function, not a lambda so it can be loaded without safe_mode=False #1262
• Update requirements and install instructions for multi-backend keras by @mattdangerw in Update requirements and install instructions for multi-backend keras #1257
• Support Keras 3 installation by @mattdangerw in Support Keras 3 installation #1258
• Remove dtensor by @mattdangerw in Remove dtensor #1268
• Add a lora dense layer by @mattdangerw in Add a lora dense layer #1263
• Factor out testing routines for models by @mattdangerw in Factor out testing routines for models #1269
• Convert T5 to Keras 3 by @nkovela1 in Convert T5 to Keras 3 #1274
• Fix missing backticks in DistilBertClassifier docstrings by @Philmod in Fix missing backticks in DistilBertClassifier docstrings #1278
• T5 checkpoint conversion with HF by @nkovela1 in T5 checkpoint conversion with HF #1277
• Use gelu_approximate directly in t5 presets by @mattdangerw in Use gelu_approximate directly in t5 presets #1284
• Add preset tests and weights URLs by @nkovela1 in Add preset tests and weights URLs #1285
• Support loading keras 3 nightly by @mattdangerw in Support loading keras 3 nightly #1286
• Remove the use of SentencePieceTrainer from tests by @tirthasheshpatel in Remove the use of SentencePieceTrainer from tests #1283
• Fix XLM-RoBERTa detokenize() by @abheesht17 in Fix XLM-RoBERTa detokenize() #1289
• Correct tie_embedding_weights and add logit checking by @nkovela1 in Correct tie_embedding_weights and add logit checking #1288
• Add detokenize testing for model tokenizers by @mattdangerw in Add detokenize testing for model tokenizers #1290
• Fix Whisper by @abheesht17 in Fix Whisper #1287
• Test against Keras 3 by @mattdangerw in Test against Keras 3 #1273
• Support TF_USE_LEGACY_KERAS by @mattdangerw in Support TF_USE_LEGACY_KERAS #1295
• Run workflows with read-only tokens by @pnacht in Run workflows with read-only tokens #1305
• Update CONTRIBUTING.md by @mattdangerw in Update CONTRIBUTING.md #1310
• Add GitHub Action for Nightly by @sampathweb in Add GitHub Action for Nightly #1309
• Fix the publish to pypi action by @mattdangerw in Fix the publish to pypi action #1311
• Fix nightly tf failure by @mattdangerw in Fix nightly tf failure #1316
• Switch deberta to use the "int" dtype by @mattdangerw in Switch deberta to use the "int" dtype #1315
• Add security policy by @pnacht in Add security policy #1319
• Fix missing export for reversible embedding by @mattdangerw in Fix missing export for reversible embedding #1327
• Add version API to keras_nlp by @grasskin in Add version API to keras_nlp #1324
• Fix Keras 3 version check by @sampathweb in Fix Keras 3 version check #1328
• Simplify running KerasNLP with Keras 3 by @mattdangerw in Simplify running KerasNLP with Keras 3 #1308
• Fix issues with version by @mattdangerw in Fix issues with version #1332
• Fix typo in whisper presets files by @mattdangerw in Fix typo in whisper presets files #1337
• ELECTRA backbone implementation in keras by @pranavvp16 in ELECTRA backbone implementation in keras #1291
• Fix t5 tokenizer expected output by @mattdangerw in Fix t5 tokenizer expected output #1348
• Add init.py for electra by @mattdangerw in Add __init__.py for electra #1352
• Remove lora dense for now by @mattdangerw in Remove lora dense for now #1359
• Adds Kokoro Build script for Keras-NLP GPU tests by @sampathweb in Adds Kokoro Build script for Keras-NLP GPU tests #1355
• Fixes GPU Test failures for Keras 3 by @sampathweb in Fixes GPU Test failures for Keras 3 #1361
• Change Continuous config to also run only large tests by @sampathweb in Change Continuous config to also run only large tests #1362
• ElectraTokenizer by @pranavvp16 in ElectraTokenizer  #1357
• Add MistralAI's 7B Transformer as a backbone in KerasNLP Models by @tirthasheshpatel in Add MistralAI's 7B Transformer as a backbone in KerasNLP Models #1314
• changing pooling output by @mbrhd in changing pooling output #1364
• Add LlamaBackbone by @shivance in Add LlamaBackbone #1203
• Align pip_build with keras by @sampathweb in Align pip_build with keras #1374
• Remove cloudbuild config by @mattdangerw in Remove cloudbuild config #1375
• Fix one last bad preset hash by @mattdangerw in Fix one last bad preset hash #1381
• Add a tokenizer for the Mistral backbone by @tirthasheshpatel in Add a tokenizer for the Mistral backbone #1383
• Kaggle Presets by @sampathweb in Kaggle Presets #1365
• Fix mistral and electra tokenizer to match kaggle changes by @mattdangerw in Fix mistral and electra tokenizer to match kaggle changes #1387
• Align requirments with Keras by @sampathweb in Align requirments with Keras #1386
• Add a preprocessor for the Mistral backbone by @tirthasheshpatel in Add a preprocessor for the Mistral backbone #1385
• Switch to always expect full Kaggle preset handles by @mattdangerw in Switch to always expect full Kaggle preset handles #1390

New Contributors

• @calvingiles made their first contribution in Make gelu a function, not a lambda so it can be loaded without safe_mode=False #1262
• @tirthasheshpatel made their first contribution in Remove the use of SentencePieceTrainer from tests #1283
• @pnacht made their first contribution in Run workflows with read-only tokens #1305
• @grasskin made their first contribution in Add version API to keras_nlp #1324
• @pranavvp16 made their first contribution in ELECTRA backbone implementation in keras #1291
• @mbrhd made their first contribution in changing pooling output #1364

Full Changelog: v0.6.4...v0.7.0

---

This discussion was created from the release v0.7.0.