Skip to content

UpstageAI/cookbook

Repository files navigation

Upstage Cookbook

This cookbook aims to highlight innovative use cases and outputs created by leveraging Upstage Solar and Document AI’s capabilities. We aim to inspire others by showcasing the versatility and excellence of Upstage features across various domains. The Upstage cookbook is open to the community. If you have an interesting idea for using APIs, spot a typo, or want to add or improve a guide, feel free to contribute!

How to use?

Get an API key from Upstage console to try examples in the cookbook. Set the environment variable UPSTAGE_API_KEY with your key, or create an .env file with the key.

Contents

  • The cookbooks/upstage directory contains official examples provided by Upstage.
  • The cookbooks/community directory contains community examples provided by any contributors.

Submission Guidelines

Submit your examples, codes, and other works to the cookbooks/community folder.

Content:

  • Source code for apps, programs, or computational projects assisted by the Upstage APIs.
  • Tutorials, educational materials, or lesson plans developed with the Upstage APIs.
  • Any other interesting applications that showcase the Upstage API's potential!

Requirements:

  • Code/text entries should be in .md, .ipynb or other common formats.
  • Include an overview describing the project/output and its purpose. Provide step-by-step instructions in a logical order and Include any relevant tips, notes, or variations.
  • Projects should be original creations or properly credited if adapted from another source.
  • Submissions should not contain explicit violating content.

Rights:

  • You maintain ownership and rights over your submitted project.
  • By submitting, you grant us permission to showcase and distribute the entry.
  • Entries cannot infringe on any third-party copyrights or intellectual property.

Let us know if you need any other guidelines or have additional criteria to include! We’re happy to continue iterating on these submission rules.

Disclaimer: The views and opinions expressed in community and partner examples do not reflect those of Upstage.

API List

API Description Example usage
Chat Build assistants using Solar Mini Chat. Link
Text Embedding Embed strings to vectors. Link
Translation Context-aware translation that leverages previous dialogues to ensure unmatched coherence and continuity in your conversations. Link
Groundedness Check Verify groundedness of assistant's response. Link
Layout Analysis Serialize documents with tables and figures. Link
Key Information Extraction Extract key information from target documents. Link
Document OCR Extract all text from any document. Link

Cookbook List

Category Notebook Description
Solar-Full-Stack LLM-101 01_hello_solar.ipynb This notebook serves as an introductory guide to the Solar-LLM series. It walks through the initial setup and basic functionalities of the Solar framework, designed to help users understand and leverage Solar for various applications. The notebook covers installation, configuration, and a simple demonstration to get users up and running with Solar.
Solar-Full-Stack LLM-101 02_prompt_engineering.ipynb This notebook explores conversational prompt engineering and question generation. It covers few-shot examples, keyword/topic generation, and Chain-of-Thought (CoT) prompting.
Solar-Full-Stack LLM-101 03_hallucinations.ipynb This notebook examines techniques to enhance the factual accuracy and truthfulness of responses from large language models (LLMs). It covers prompt engineering methods like few-shot examples, chain-of-thought prompting, and keyword prompting to prevent hallucinations.
Solar-Full-Stack LLM-101 04_CAG_GC.ipynb This notebook shows how to use Solar and Langchain for answering questions based on a given context and checking the groundedness (factual accuracy) of the answers using the Upstage Groundedness Check API.
Solar-Full-Stack LLM-101 05_1_ChromaDB.ipynb Get how to create a vector store using the Chroma vectorstore and Upstage Embeddings for semantic search. This allows for efficient storage and retrieval of embedded document vectors for semantic search queries.
Solar-Full-Stack LLM-101 05_2_MongoDB.ipynb How to set up a MongoDB Atlas cluster and create vector search indexes for semantic search. It shows how to use the MongoDBAtlasVectorSearch class from LangChain to store documents and their embeddings in MongoDB, and perform hybrid searches that combine keyword-based and vector-based retrieval techniques, with results ranked using reciprocal rank fusion.
Solar-Full-Stack LLM-101 05_3_OracleDB.ipynb This notebook shows how to use the OracleVectorDB library to store and retrieve text data using vector embeddings. It showcases splitting text into chunks, creating vector embeddings, and indexing/searching the embedded text data.
Solar-Full-Stack LLM-101 06_PDF_CAG.ipynb Check how to analyze PDF documents using Solar. It includes steps to extract text from PDFs, process the text using various natural language processing techniques, and generate comprehensive analyses.
Solar-Full-Stack LLM-101 07_LA_CAG.ipynb Explore how to perform layout analysis on documents using a Solar. It includes techniques for detecting and interpreting the structural elements of a document, such as headings, paragraphs, tables, and figures. And chat based on result.
Solar-Full-Stack LLM-101 08_RAG.ipynb Learn how to implement Retrieval Augmented Generation (RAG). It showcases how to load documents, create a retriever, and use it with an Solar to answer questions based on retrieved relevant information.
Solar-Full-Stack LLM-101 09_Smart_RAG.ipynb Explore how to use LangChain to build a Smart RAG system that can answer questions either from a given context or by searching the internet if the context is insufficient. It utilizes the UpstageEmbeddings model, the Tavily search API, and conditional logic to determine if a question can be answered from the context. If not, it searches Tavily and augments the context before providing the final answer.
Solar-Full-Stack LLM-101 10_tool_RAG.ipynb Learn how to building a Tool RAG system using LangChain, which can answer queries by intelligently selecting and utilizing different tools or data sources.
Solar-Full-Stack LLM-101 11_summary_writing_translation.ipynb This notebook shows using the LangChain library with the Upstage Embeddings model for various natural language processing tasks such as summarization, text simplification, and translation. It showcases how to create prompts and chains for summarizing text, rewriting text to be more understandable, translating English to Korean, and providing translations in a specific style by conditioning on examples.
LangGraph-Self-RAG langgraph_self_rag.ipynb This notebook provides a comprehensive guide on implementing Self Retrieval-Augmented-Generation (Self-RAG) using LangGraph. It demonstrates how to enhance text generation by integrating self-retrieval mechanisms, allowing the model to dynamically fetch relevant information to improve accuracy and relevance.
function_calling.ipynb This notebook provides the most basic way to use function calling.
llamaindex_rag.ipynb In this notebook provides a comprehensive way to build a Retrieval-Augmented-Generation using the llama-index Upstage integration package.