Support Llama API as a Chat Generator

[seyeong-han]

trafficstars

This pull request introduces the initial implementation of the meta-llama-haystack integration, enabling the use of Meta's Llama API for text generation within the Haystack framework. The changes include the addition of the core LlamaChatGenerator component, example usage, project configuration, and documentation.

Core Features and Implementation:

• LlamaChatGenerator Component:

  • Added LlamaChatGenerator class to enable text generation using Meta's Llama API. This class extends OpenAIChatGenerator and supports features like streaming responses, customizable generation parameters, and integration with the Haystack framework. (integrations/meta_llama/src/haystack_integrations/components/generators/llama/chat/chat_generator.py)

• Example Usage:

  • Added a Python example demonstrating how to create a Retrieval-Augmented Generation (RAG) pipeline using the LlamaChatGenerator and Haystack components. (integrations/meta_llama/examples/rag_with_llama.py)

Project Setup and Configuration:

• Project Metadata:

  • Added pyproject.toml to define project metadata, dependencies, and build configuration for the meta-llama-haystack package. (integrations/meta_llama/pyproject.toml)

• Documentation and Changelog:

  • Added README.md with installation instructions, API overview, and licensing information. (integrations/meta_llama/README.md)
  • Added CHANGELOG.md for tracking changes. (integrations/meta_llama/CHANGELOG.md)

• Documentation Generation:

  • Added pydoc/config.yml for generating API documentation using haystack-pydoc-tools. (integrations/meta_llama/pydoc/config.yml)

[CLAassistant]

All committers have signed the CLA.

[bilgeyucel]

Thank you for opening this PR @seyeong-han ! It looks quite good to me already. My first main point is about aligning the implementation more with the most recently added OpenRouterChatGenerator, which also inherits from OpenAIChatGenerator and is very similar. The changes I suggest are about the streaming callback and the tools. And the second point is about the name of the component and the module. I thought about users importing it and given they installed the integration meta-llama-haystack, it might be more intuitive to have from haystack_integrations.components.generators.meta_llama import LlamaChatGenerator or from haystack_integrations.components.generators.meta_llama import MetaLlamaChatGenerator instead of from haystack_integrations.components.generators.llama import LlamaChatGenerator. I'll ask another team member for their opinion.

I agree with @julian-risch's comment here. Between suggestions, I think from haystack_integrations.components.generators.meta_llama import MetaLlamaChatGenerator fits better to the Haystack codebase and is more straightforward for users to understand, as it has the same prefix as the package name and represents both the company and API name

[seyeong-han]

@julian-risch @bilgeyucel Thanks for all your review!

We'll add additional Agent example after merging this PR :)

[seyeong-han]

@julian-risch I signed the CLA using my meta email but it hasn't been confirmed. Should I have typed my representative email in Github?

[julian-risch]

Looks good to me! 👍 I'll make some additions that we need for a new integration. After that, I'll approve, merge, and release.

• [x] A new label named like integration:<your integration name> has been added to the list of labels for this repository
• [x] There is a Github workflow running the tests for the integration nightly and at every PR
• [x] The labeler.yml file has been updated
• [x] Add to README on the repository overview page

[seyeong-han]

Thanks @julian-risch ! I'll let you know after adding the meta_llama documentation in haystack-integrations repository.