Releases: caikit/caikit-nlp
Releases Β· caikit/caikit-nlp
v0.4.4 - Embeddings with sentence transformers
What's Changed
- Embeddings: IPEX optimize requires model.eval() called first. by @markstur in #318
- Revert "Add granite modeling llama main" by @gkumbhat in #321
- Add sentence-transformers to dependencies for real by @markstur in #319
- added tokenization tasks to serve tokenization requests by @swith004 in #325
- Wrapped sentence transformer by @markstur in #328
Full Changelog: v0.4.3...v0.4.4
v0.4.3 - Support for gpt megatron architecture
What's Changed
- bump dockerfile ubi base by @dtrifiro in #303
- π₯ Handle empty text for filtered span classification by @evaline-ju in #304
- π Add model and tokenizer time logging by @gkumbhat in #309
- Add granite modeling llama main by @gkumbhat in #314
- Embeddings: Fix retry error handling to return first exception. Default to zero retries. by @markstur in #316
Full Changelog: v0.4.2...v0.4.3
v0.4.2
What's Changed
- set learning_rate typing to float by @alonh in #300
- Update caikit[runtime-grpc,runtime-http] requirement from <0.26.0,>=0.25.0 to >=0.25.0,<0.27.0 by @dependabot in #301
New Contributors
Full Changelog: v0.4.1...v0.4.2
v0.4.1 - first pypi release
0.4.0 - text embeddings
What's Changed
- tests: make models fixtures session-scoped by @dtrifiro in #262
- π₯ Disallow empty train streams by @evaline-ju in #259
- Add dockerfile by @dtrifiro in #244
- Add preserve input text inference by @Ssukriti in #267
- π· Update github action workflow rule for release branches by @gkumbhat in #269
- Add embedding task by @markstur in #224
- Add rerank and sentence-similarity tasks to text embedding module by @markstur in #235
- π Fix the ordering of arguments in all the stream text gen functions by @gkumbhat in #276
- π§βπ» Improve regex for directory validation by @gkumbhat in #275
- Text Embedding: Add optimizations with PT2_COMPILE and/or IPEX by @markstur in #274
- Text Embedding: Truncation allow or fail with optional parameter by @markstur in #277
- Add Sukriti by @gkumbhat in #286
- dockerfile: build wheel in container by @dtrifiro in #289
- Migrate hf trainer by @gkumbhat in #287
- Add max size information to error message by @gkumbhat in #290
- Fix prompt unload (main) by @joerunde in #284
- fix example:pass learning rate to peft by @Ssukriti in #294
- Text Embedding: Fix concurrency errors by @markstur in #291
- π Tox install by @evaline-ju in #296
- Revert "Migrate hf trainer" by @Ssukriti in #297
New Contributors
Full Changelog: 0.3.0...0.4.0
0.3.3 - TGIS prompt purge bugfix
Fixes a bug where the code to unload prompt artifacts on model deletion was calling an api that didn't exist
0.3.2 - Error message improvements
Includes #281
0.3.1
0.3.0
What's Changed
- Expose seed by @tharapalanivel in #201
- Refactor / Consolidate Batch Tokenization for Causal Language Modeling by @alex-jw-brooks in #192
- Validation checks by @tharapalanivel in #200
- π Align inference params order by @tharapalanivel in #204
- fix link for contributing info by @jimknochelmann in #195
- fix imports for caikit >=0.15.0 by @dtrifiro in #209
- Fix tokenization path by @gkumbhat in #202
- Add ft loss logging by @gkumbhat in #208
- β¨ Add support for reading base model from base models dir by @gkumbhat in #211
- use arg use_cpu instead of no_cuda by @alex-jw-brooks in #215
- Remove stop sequence default by @tharapalanivel in #213
- π Add support for setting random seed for prompt tuning training by @gkumbhat in #217
- Causal LM tokenization: Chunking and seq2seq Forward by @alex-jw-brooks in #206
- Resync inf params tgis by @gkumbhat in #220
- π₯ Disallow empty train streams by @evaline-ju in #223
- β¬οΈ Update caikit[runtime-grpc,runtime-http] requirement from <0.21.0,>=0.18.1 to >=0.18.1,<0.22.0 by @dependabot in #227
- model_run_utils: return correct input_token_count by @dtrifiro in #228
- Add support for sequence lengths in eval by @alex-jw-brooks in #232
- Revert ":goal_net: Disallow empty train streams" by @gkumbhat in #234
- β¬οΈ Update build requirement from <1.0,>=0.10.0 to >=0.10.0,<2.0 by @dependabot in #164
- Make verbalizer rendering optional in eval script by @alex-jw-brooks in #231
- π§βπ» Fix docstring formatting function call by @gkumbhat in #219
- proposal: use setuptools/setuptools_scm for building by @dtrifiro in #216
- Update to tasks by @tharapalanivel in #236
- fix pylint warnings by @dtrifiro in #240
- pyproject: use setuptools' automatic discovery by @dtrifiro in #239
- ππ§ Fix sequencing of arguments on run function by @gkumbhat in #241
- tox: run pytest with --durations flag to list tests run time by @dtrifiro in #246
- return correct finish reasons in generate_text_func by @dtrifiro in #210
- Speed up tests by @dtrifiro in #249
- π Fix loss tensor to float conversion by @gkumbhat in #258
- Add data limitation by @gkumbhat in #255
- fix uncaught exception in http client by @dtrifiro in #247
- Update training data validation to consider global and module level defaults by @gkumbhat in #264
New Contributors
- @jimknochelmann made their first contribution in #195
- @dtrifiro made their first contribution in #209
Full Changelog: 0.2.1...0.3.0
0.2.1
What's Changed
- Added benchmarking results to main README.md by @olson-ibm in #176
- Experiment sharding rebase (multi GPU support for Find Tuning) by @alex-jw-brooks and @gkumbhat in #156
- π Fix eval script for updated run datamodel by @gkumbhat in #184
- Add loss logging prompt-tuning by @gkumbhat in #189
- fix: use cli args for metrics to run by @timrbula in #193
- [caikit-nlp-163] Refactor peft module to take out common peft config functionality by @rawkintrevo and @alex-jw-brooks in #197
- SupportCaikitZeroTwenty: Bump caikit range to allow for 0.20 by @gabe-l-hart in #199
- Fix dtype prompt tuning by @gkumbhat in #198
New Contributors
Full Changelog: 0.2.0...0.2.1