v0.4.1

What's Changed

🐛 Bug fixes

• Correctly handle file responses
• Use tqdm.auto instead of tqdm.tqdm to improve display in Jupyter Notebooks

Full Changelog: v0.4.0...v0.4.1

v0.4.0

What's Changed

⚠️ Switch to Pydantic V2

• In case your application is dependent on Pydantic V1, refer to the migration guide.
• If you cannot upgrade, stick to the previous version 0.3.2.

Full Changelog: v0.3.2...v0.4.0

v0.3.2

What's Changed

🐛 Bug fixes

• Correctly handle async errors and process abortion

🔧 Configuration Changes

• Increase async generate/tokenize retry limits from 3 to 5

Full Changelog: v0.3.1...v0.3.2

v0.3.1

What's Changed

🚀 Features / Enhancements

• Handle concurrency limits for generate and generate_as_completed methods.
• Add automatic handling of rate limits for the tokenize endpoint (tokenize_async method).
• Added stop_sequence parameter for generated output (non-empty token which caused the generation to stop) + added - include_stop_sequence parameter for the GenerateParams (it indicates whether the stop sequence (which caused the generation to stop) is part of the generated text. The default value depends on the model in use).
• Removed hidden stop_sequences removal inside the LangChainInterface, which can now be controlled via the include_stop_sequence parameter.
• Improve general error handling + method signatures (improve Python typings).

🐛 Bug fixes

• Fix stacked progress bar (generate_async method)
• Handle cases when the package is used inside the asyncio environment
• Hide warning when an unknown field is retrieved in the generated response

Full Changelog: v0.3.0...v0.3.1

v0.3.0

What's Changed

🚀 Features / Enhancements

• Added Hugging Face Agent support; see an example.
• Drastically improve the speed of generate_async method - the concurrency limit is now automatically inferred from the API. (custom setting of ConnectionManager.MAX_CONCURRENT_GENERATE will be ignored). In case you want to slow down the speed of generating, just pass the following parameter to the method: max_concurrency_limit=1 or any other value.
• Increase the default tokenize processing limits from 5 requests per second to 10 requests per second (this will be increased in the future).

🐛 Bug fixes

• Throws on unhandled exceptions during the generate_async calls.
  Correctly cleanups the async HTTP clients when the task/calculation is being cancelled (for instance, you call generate_async in Jupyter - Notebook and then click the stop button). This should prevent receiving the Can't have two active async_generate_clients error.
• Fix async support for newer LangChain versions (>=0.0.300)
• Fix LangChain PromptTemplate import warning in newer versions of LangChain
• Correctly handle server errors when streaming
• Fix tune_methods method

v0.2.8

What's Changed

🚀 Features / Enhancements

• Added moderation support; now you can retrieve HAP for generated requests (example)
• Internally improve streaming processing (poor or unstable internet connection)
• Internally improve server response parsing and error handling
• Add a user-agent header to distinguish Python SDK on the API

🐛 Bug fixes

• LangChain - correct handling of stop_sequences
• Correctly set versions of used dependencies (httpx / pyyaml)
• Prevents unexpected modifications to user's GenerateParams passed to the Model class
• Prevents unexpected errors when GenerateParams contains stream=True and generate (non-stream) version is called

🔧 Configuration changes

• Remove API version from the API endpoint string

Full Changelog: v0.2.7...v0.2.8

v0.2.7

What's Changed

• feat(langchain) - generate method by @Tomas2D in #157
• fix(params): do not strip special characters by @Tomas2D in #153
• fix: correct httpx dependency version by @Tomas2D in #158

Full Changelog: v0.2.6...v0.2.7

v0.2.6

What's Changed

• feat(langchain): add streaming support by @Tomas2D in #144
• feat(http): allow override httpx options by @Tomas2D in #149
• feat: add typical_p parameter by @Tomas2D in #135
• chore: update examples by @Tomas2D in #136
• docs: mention CLI in README by @Tomas2D in #143
• chore: adding escapting of backslashes for re.sub value by @assaftibm in #84
• chore: correct README.md typo by @ind1go in #148
• update schema for stop_sequences generate param by @mirianfsilva in #142

New Contributors

• @assaftibm made their first contribution in #84
• @ind1go made their first contribution in #148

Full Changelog: v0.2.5...v0.2.6

v0.2.5

• TOUs handling
• Update Pydantic version
• Update examples

Full Changelog: v0.2.4...v0.2.5

v0.2.4

Updated the documentation (imports of credentials)
Updated schemas for config
Added params in GeneratedParams
Updated examples
Updated tests