Basic RAG with LangChain and LangGraph - Ollama

Explore the 1 billion parameter llama model, detailing eight-bit quantization, instruction-tuned architecture, 32 attention networks across 16 decoder blocks, and template-driven tool and function calling in LangChain and LangGraph.

copy and rename a model with Ollama cp, run it from a copy, chat with it in chat mode, check ps, and stop with Ollama stop, noting the five-minute offload.

Create and run an Ollama model from a predefined model file using the create command with model file.txt, loading from llama 3.2 and setting temperature, context, and a system message.

Learn to use the llama model command to load llama 3.21 billion model and compare with the Seldon model, view inside commands, and manage parameters, system messages, and templates.

Learn to use Ollama's set, save, and load commands to configure system messages, enable or disable history, store and load models, and observe verbose diagnostics.

Discover how to interact with the Ulama raw API, using generate and chat endpoints, and learn to configure model, prompt, system messages, and context for effective outputs.

Learn to load uncensored models from Hugging Face into llama, download an older uncensored model, and run educational prompts to explore capabilities while noting legal and ethical restrictions.

Explore generating uncensored content in LangChain by switching between Seldon, Sherlock, and uncensored models, and learn to stream token-by-token outputs for educational purposes.

Track and analyze LLM inputs and outputs in Lang Smith, viewing runs, latency, token usage, and model metadata; use Lang Fusion or Opic for private data.

Explore the line graph in the Lang graph library by reviewing GitHub repository, its APIs, and memory design using type hinting that stores agent state with nodes and tool nodes.

Explore the LangGraph line graph module, and learn how to import graph components from the line graph package and its memory checkpoints, while understanding why the GitHub repository context matters.

Explore OpenAI and the Llama Vision model to describe images in PDFs for RAG workflows. Compare local Llama Vision with OpenAI API for faster image descriptions and setup.

Load the OpenAI API key from a dot env file and compare private LM with the llama local LM and OpenAI chat GPT-4 for generating image descriptions.

Explore converting a pdf to markdown with doc link to improve LLM understanding, compare doc link and markdown options, and integrate with LangChain doc link loader.

Learn to convert online documents to markdown with LangChain doc link loader and document loader, configure export types, and manage local paths for seamless markdown output.

Use os.walk to traverse directories, filter for PDF, Microsoft documents, PPT, and Excel files, and convert them into structured markdown outputs stored in a target directory.

Discover how to extract images from pdf or docx documents, describe them with vision models, and convert image data into text or tabular form for extraction.

Learn to extract images from a pdf, convert figures to markdown with embedded or referenced images, and generate text descriptions using a language model.

Split markdown data page by page with LangChain's markdown header splitter to isolate each page's content, then pass content and images to an LLM for per-page image descriptions.

Compare enriched markdown with bare markdown by showing image enrichment, extracted image descriptions, and producing a perfect markdown file for lm-driven rag data workflows.

Generate text embeddings using the gnomic embedded text model from llama via ollama. Store vectors in an in memory doc store and index them for ingestion into a vector database.

Learn how to retrieve documents from a FAISS vector store using embeddings, cosine similarity, and MMR to fetch top relevant chunks, and apply filters to target specific companies.

Build an agentic rag that ingests and retrieves documents in a vector store, then uses an LLM and a retriever to generate answers, rewriting queries when needed.

Implement an agent node that uses the graph state to decide retrieval via the retriever tool or end, then generate or rewrite the answer with the analyst tool.

Watch how an agent retrieves documents, checks relevance, and decides to rewrite or generate a final answer, streaming output in three-sentence bursts and comparing across queries like Facebook and Google.