Interactive Graphene Documentation
Overview
Tutorials are crucial for seamlessly integrating new implementations and adding essential features. As the Eclipse Graphene platform expands, providing a wealth of tutorials becomes necessary for users navigating its intricacies. In such scenarios, we may encounter documentation challenges such as inadequacy and inaccessibility to understand the tutorials.
Leveraging Large Language Models (LLMs) offers a solution for the laborious documentation process. Integrating LLMs and LangChain enables the automated generation of comprehensive documentation for the platform's tutorials folder, thus streamlining the entire process. This approach aims to ensure well-structured, informative content, addressing the evolving needs of users and stakeholders.
How it Works: Proposed Workflow:
Leveraging LLMs
The core intent is to employ Large Language Models (LLMs), pre-trained models trained on humongous amounts of text data, to automate the document generation process. These LLMs can be either publicly available or closed-source LLMs, which are proprietary models. The shift from manual document authoring to leveraging LLMs improves efficiency significantly, reducing labor intensity.The following illustrations depict the workflow for both open-source and closed-source LLMs.
OpenAI's LLMs | Open-source LLMs |
---|---|
Incorporating LangChain
LangChain is a sophisticated framework meticulously composed to streamline the document generation process, utilizing the capabilities of LLMs. This framework simplifies the orchestration and integration of LLMs for the specific purpose of document generation. It provides vital concepts such as custom prompt templates, chains, and interfaces to access LLMs, all of which contribute to an efficient and structured workflow. It thus offers a robust framework for seamlessly incorporating hybrid AI into production pipelines.- Integration of LangChain components -
- DirectoryLoader
- PromptTemplate
- Models I/O
- Chains
- Agent Tooling
- Lang Chain Expression Language (LCEL)
Tasks
-
Formulation of the LLM Pipeline -
Conduct a comprehensive literature survey leveraging the capabilities of the LangChain framework -
Investigate the potential of OpenAI LLMs -
Explore MistralAI and LLama-2 LLMs thoroughly -
Integrate LLMs seamlessly with the LangChain platform -
Deploy MistralAI and LLama-2 LLMs on Fraunhofer Edge Cloud (FEC) -
Experimental LLM pipeline design: v1.0, v1.1 -
Refine and finalize the LLM pipeline for optimal performance -
Jupyter Notebook implementation for both OpenAI and MistralAI LLM pipelines -
Containerization of the various components in the Eclipse Graphene platform for the pipeline
Containerization
Three containers are being developed to implement the LLM pipeline on the Graphene platform. These specific Docker containers are designed with a generic approach, making them adaptable for future use in diverse applications, focussing on reusability across different LLM pipelines.
container_openai
This Docker container hosts the OpenAI API's model selection process required for selecting the LLMs on the Graphene platform. It encapsulates the necessary API parameters, allowing users to select by inputting their chosen parameters.
container_readme_gen_module
This container dynamically produces README files tailored to the selected Graphene tutorials. It utilizes the LangChain Expression Language (LCEL) for summarization and showcases the array of Agents and Tools within LangChain to generate README.md files specific to the chosen tutorial.
container_user_ratings
This Docker container manages the generated README file's user ratings and feedback mechanisms.
Other tasks
-
Onboard the model to the Dev AI-Builder. Provide a comprehensive pipeline description, instructing users to generate a document on the homepage. -
Test various tutorials to assess and enrich the user interactivity of the pipeline, focusing on refining the README.
References
- https://platform.openai.com/docs/models/gpt-3-5
- https://huggingface.co/meta-llama/Llama-2-13b-hf
- https://www.langchain.com/
Please refer to the following document for a detailed overview of the literature survey, older pipeline versions, and other related information. IGD_Doc