Skip to content

Latest commit

 

History

History
158 lines (92 loc) · 6.28 KB

CHANGELOG.md

File metadata and controls

158 lines (92 loc) · 6.28 KB
Raw

Changelog

Current version: 0.0.35

PyPi link

0.0.35 - October 22, 2024

Added

Changed

  • claude-3.5-sonnet now points to version claude-3-5-sonnet-latest

0.0.34 - September 30, 2024

Caution

This release has breaking changes! Please read the changelog carefully.

Added

  • New supported models gemma-2-9b, llama-3.2-1b, and llama-3.2-3b via Groq.

Changed

  • In order to be more consistent with l2m2's naming scheme, the following model ids have been updated:
    • llama3-8b → llama-3-8b
    • llama3-70b → llama-3-70b
    • llama3.1-8b → llama-3.1-8b
    • llama3.1-70b → llama-3.1-70b
    • llama3.1-405b → llama-3.1-405b
  • This is a breaking change!!! Calls using the old model_ids (llama3-8b, etc.) will fail.

Removed

  • Provider octoai has been removed as they have been acquired and are shutting down their cloud platform. This is a breaking change!!! Calls using the octoai provider will fail.
    • All previous OctoAI supported models (mixtral-8x22b, mixtral-8x7b, mistral-7b, llama-3-70b, llama-3.1-8b, llama-3.1-70b, and llama-3.1-405b) are still available via Mistral, Groq, and/or Replicate.

0.0.33 - September 11, 2024

Changed

  • Updated gpt-4o version from gpt-4o-2024-05-13 to gpt-4o-2024-08-06.

0.0.32 - August 5, 2024

Added

  • Mistral provider support via La Plateforme.

  • Mistral Large 2 model availibility from Mistral.

  • Mistral 7B, Mixtral 8x7B, and Mixtral 8x22B model availibility from Mistral in addition to existing providers.

  • 0.0.30 and 0.0.31 are skipped due to a packaging error and a model key typo.

0.0.29 - August 4, 2024

Caution

This release has breaking changes! Please read the changelog carefully.

Added

  • alt_memory and bypass_memory have been added as parameters to call and call_custom in LLMClient and AsyncLLMClient. These parameters allow you to specify alternative memory streams to use for the call, or to bypass memory entirely.

Changed

  • Previously, the LLMClient and AsyncLLMClient constructors took memory_type, memory_window_size, and memory_loading_type as arguments. Now, it just takes memory as an argument, while window_size and loading_type can be set on the memory object itself. These changes make the memory API far more consistent and easy to use, especially with the additions of alt_memory and bypass_memory.

Removed

  • The MemoryType enum has been removed. This is a breaking change!!! Instances of client = LLMClient(memory_type=MemoryType.CHAT) should be replaced with client = LLMClient(memory=ChatMemory()), and so on.

0.0.28 - August 3, 2024

Added

  • Providers can now be activated by default via the following environment variables:
    • OPENAI_API_KEY for OpenAI
    • ANTHROPIC_API_KEY for Anthropic
    • CO_API_KEY for Cohere
    • GOOGLE_API_KEY for Google
    • GROQ_API_KEY for Groq
    • REPLICATE_API_TOKEN for Replicate
    • OCTOAI_TOKEN for OctoAI

0.0.27 - July 24, 2024

Added

  • OctoAI provider support.
  • Llama 3.1 availibility, in sizes 8B (via OctoAI), 70B (via OctoAI), and 405B (via both OctoAI and Replicate).
  • Mistral 7B and Mixtral 8x22B via OctoAI.
  • LLMOperationError exception, raised when a feature or mode is not supported by a particular model.

Fixed

  • Rate limit errors would sometimes give the model id as None in the error message. This has been fixed.

0.0.26 - July 19, 2024

Added

0.0.25 - July 12, 2024

Added

  • Custom exception LLMRateLimitError, raised when an LLM call returns a 429 status code.

0.0.24 - July 11, 2024

Added

  • The ability to specify a custom timeout for LLM calls by passing a timeout argument to call or call_custom (defaults to 10 seconds).
  • A custom exception LLMTimeoutError which is raised when an LLM call times out, along with a more helpful message than httpx's default timeout error.

Fixed

  • Calls to Anthropic with large context windows were sometimes timing out, prompting this change.

0.0.23 - June 30, 2024

Fixed

  • Major bug where l2m2 would cause environments without typing_extensions installed to crash due to it not being listed as an external dependency. This has been fixed by adding typing_extensions as an external dependency.

Changed

  • This bug wasn't caught becuase integration tests were not running in a clean environment – (i.e., typing_extensions was already installed from one of the dev dependencies). To prevent this from happening again, I made make itest uninstall all Python dependencies before running.

0.0.22 - June 22, 2024

Fixed

  • In 0.0.21, async calls were blocking due to the use of requests. 0.0.22 replaces requests with httpx to allow for fully asynchoronous behavior.

Changed

  • AsyncLLMClient should now be instantiated with a context manager (async with AsyncLLMClient() as client:) to ensure proper cleanup of the httpx client.
  • In AsyncLLMClient, call_async and call_custom_async have been renamed to call and call_custom respectively, with asynchronous behavior.

Removed

  • call_concurrent and call_custom_concurrent have been removed due to unnecessary complexity and lack of use.

0.0.21 - June 20, 2024

Added

  • This changelog (finally – oops)
  • Support for Anthropic's Claude 3.5 Sonnet released today

Changed

  • L2M2 is now fully HTTP based with no external dependencies, taking the total recursive dependency count from ~60 to 0 and massively simplifying the unit test suite.
  • Non-native JSON mode strategy now defaults to prepend for Anthropic models and strip for all others.