v0.7.2: Support for Python 3.12 and a Torch-related bugfix

✨ New features and improvements

• Support for Python 3.12 (#466).

🔴 Bug fixes

• Ensure that Torch is not imported or used unless it is available (#467).

⚠️ Backwards incompatibilities

No backwards incompatibilities w.r.t. v0.7.1.

👥 Contributors

@honnibal, @ines, @magdaaniol, @svlandeg

v0.7.1: Update `langchain` integration; fix EL task, model endpoints

🔴 Bug fixes

• Remove misguided check for fixed endpoints in OpenAI REST implementation (#429)
• Fix bug in EL task if non-highlighted entities are in text (#375)
• Update langchain integration to support langchain >= 0.1, < 0.2 (#433)

⚠️ Backwards incompatibilities

No backwards incompatibilities w.r.t. v0.7.0.

📖 Documentation and examples

• Link task mentions in readme to docs (#421)

👥 Contributors

@honnibal, @ines, @magdaaniol, @rmitsch, @svlandeg

v0.7.0: Supporting arbitrarily long docs and various new tasks

✨ New features and improvements

• NEW: Support arbitrarily long docs leveraging map-reduce and document sharding. Read more here (#332)
• NEW: New tasks -
  • EntityLinkingTask: use LLMs to link entity mentions in texts to entities in a custom knowledge base. Read more here (#241)
  • TranslationTask: translate documents from and to arbitrary languages. Read more here (#396)
  • RawTask: prompt models with only doc content, no prompt scaffolding, for a maximum of flexibility. Read more here (#395)
• Allow custom endpoints for all models to support usage of fine-tuned models (#390)

🔴 Bug fixes

• Fix issues with previously incomplete integration of LangChain objects (#417)

⚠️ Backwards incompatibilities

No backwards incompatibilities w.r.t. v0.6.x.

📖 Documentation and examples

• Descriptions of EntityLinkingTask, TranslationTaskand RawTask (explosion/spaCy#12988, explosion/spaCy#13183, explosion/spaCy#13180)
• Introduction to and description of doc sharding/map-reduce for the support of infinitely long docs (explosion/spaCy#13214)

👥 Contributors

@honnibal, @ines, @rmitsch, @svlandeg

v0.6.4: Fix `langchain` model init

🔴 Bug fixes

• Fix langchain model init for models not expecting model_name as model ID arg (#374)
• Ignore Pydantic deprecation warnings stemming from torch (#375)

⚠️ Backwards incompatibilities

No backwards incompatibilities w.r.t. v0.6.0.

👥 Contributors

@honnibal, @ines, @shadeMe, @rmitsch

v0.6.3: Improved device handling for HF models, new OpenAI models, various fixes, etc.

✨ New features and improvements

• Scoring support for sentiment task (#331)
• Add label definitions for LabeledTasks (#340)
• Support langchain 0.0.335 (#365)
• Support new OpenAI models by allowing arbitrary model names for existing model families (#356)

🔴 Bug fixes

• Fix REL prompt generation outputting internal IDs instead of string values for labels (#367)
• Fix device handling for HF models, allow passing torch_dtype in config (#359)
• Fix incorrect base URL for Azure OpenAI models (#337)

⚠️ Backwards incompatibilities

No backwards incompatibilities w.r.t. v0.6.0.

👥 Contributors

@habibhaidari1, @honnibal, @ines, @shadeMe, @rmitsch, @svlandeg, @viveksilimkhan1

v0.6.2: Patch anthropic-version header

🔴 Bug fixes

• Anthropic requests include a anthropic-version header (instead of anthropic_version) (#328)

⚠️ Backwards incompatibilities

No backwards incompatibilities w.r.t. v0.6.0.

👥 Contributors

@honnibal, @ines, @rmitsch

v0.6.1: Fix model initialization for Llama 2 and requests for Anthropic

🔴 Bug fixes

• Llama 2 init doesn't use use_auth_token=True by default anymore (#325)
• Anthropic requests include a anthropic_version header (#325)

Both bugs couldn't be reproduced on our end, so we can't guarantee that these bugs are fixed now. #323 and #324 for a follow-up.

⚠️ Backwards incompatibilities

No backwards incompatibilities w.r.t. v0.6.0.

👥 Contributors

@honnibal, @ines, @rmitsch

v0.6.0: PaLM, Azure OpenAI, Mistral & fixed OS model responses

✨ New features and improvements

• NEW: Native REST support for...
  • PaLM (#305)
  • Azure OpenAI (#316)
  • Mistral 7B (#313)
  • OpenAI gpt-3.5-turbo-instruct - supports batching! (#300)
• Support langchain==0.0.302 (#308)

🔴 Bug fixes

• Hugging Face models return just the prompt response instead of prompt + prompt response (#315)

⚠️ Backwards incompatibilities

No backwards incompatibilities w.r.t. v0.5.x.

📖 Documentation and examples

• New usage example for spacy.NER.v3 with Dolly (#302)

👥 Contributors

@honnibal, @ines, @rmitsch

v0.5.1: Fix `confection` dependency pin

🔴 Bug fixes

• Fix Fix confection dependency pin (#290)

v0.5.0: Improved user API and novel Chain-of-Thought prompting for more accurate NER

✨ New features and improvements

• NEW: More accurate Chain-of-Thought (CoT) NER prompting with spacy.NER.v3(#180)
• NEW: Task-specific component factories for llm_ner , llm_spancat, llm_rel,llm_textcat , llm_sentiment, llm_summarization(#243, #283)
• NEW: Implementation of add_label functionality to more easily work with an llm component directly in Python code and not (necessarily) through the config system (#277)
• New v2 model versions for the OpenAI models that set reasonable defaults for temperature and max_tokens (#236)
• Functionality to ignore occassional blocking errors from Cohere (#233)
• Support for Pydantic v1 and v2 (#261, #275)
• Internal refactoring, including renaming of v1 Jinja templates (#242)
• Empty the cache of torch.cuda in between calls (#242)
• Various improvements to the test suite and CI

🔴 Bug fixes

• Fix Anthropic chat endpoints (#230)

⚠️ Backwards incompatibilities

• Though significant refactoring of internal modules has happened, this release should not introduce any backwards incompatibilities for user-facing functionality.
• Check our migration guide if you want to update the SpanCat or NER task from v1 or v2 to v3.

📖 Documentation and examples

• Updated usage documentation
• Updated API documentation
• New Chain-of-Though example with GPT 3.5
• New textcat example with Dolly

👥 Contributors

@adrianeboyd, @honnibal, @ines, @kabirkhan, @ljvmiranda921, @rmitsch, @svlandeg, @victorialslocum, @vinbo8