Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Building Performant AI Applications with Nvidia...

Tuana Çelik
September 05, 2024
39

Building Performant AI Applications with Nvidia NIMs and Haystack

Tuana Çelik

September 05, 2024
Tweet

Transcript

  1. Introduction to inferencing with Mark 1. Intro to NIMs with

    Mark 2. LLM NIMs, Embedding NIMs and Re- ranking NIMs a. Building RAG with NIMs and Haystack, with Tuana 3. Deploy and Scale with Anshul 4. Q&A 5. Agenda @TUANACELIK /IN/TUANACELIK
  2. FAQ The code and examples we show are all available

    on GitHub Yes this is recorded and will be up in the Haystack YouTube channel ASAP
  3. RAG Question Answering Given the documents, answer the question. Documents:

    {{ documents}} Question: {{ question }} @TUANACELIK /IN/TUANACELIK
  4. RAG Question Answering Given the documents, answer the question. Documents:

    {{ documents}} Question: {{ question }} Question @TUANACELIK /IN/TUANACELIK
  5. RAG Question Answering Given the documents, answer the question. Documents:

    {{ documents}} Question: {{ question }} Question @TUANACELIK /IN/TUANACELIK
  6. RAG Question Question Answering Given the documents, answer the question.

    Documents: Question: {{ question }} @TUANACELIK /IN/TUANACELIK
  7. RAG Embed Query Build Prompt Query Retrieve Document(s) Prompt Query

    Embedding Embeddings Documents Given the documents, answer the question. Documents: {{ documents}} Question: {{ query }} Query Prompt Prompt Response Generate Response @TUANACELIK /IN/TUANACELIK
  8. RAG Embed Query Build Prompt Query Retrieve Document(s) Prompt Query

    Embedding Embeddings Documents Answer the question given the context. Question: {{ query }} Context: {% for document in documents %} {{ document.content }} {% endfor %} Answer: Query Prompt Prompt Response Generate Response @TUANACELIK /IN/TUANACELIK meta/llama-3.1-70b-instruct snowflake/arctic-embed-l
  9. CONDITIONAL ACTIONS SEARCH THE WEB IF... ConditonalRouter WebSearch go_to_websearch answer

    Answer Generation Retrieval Prompt Response @TUANACELIK /IN/TUANACELIK
  10. Fully open-source framework built in Python for custom AI applications

    Provides tools that developers need to build and customize state-of-the-art AI systems Building blocks: Pipelines & Components Component Component Pipeline @TUANACELIK /IN/TUANACELIK
  11. Haystack 2.0: DAG Pipeline 📚 Announced March 11, 2024 Component

    Component Component @TUANACELIK /IN/TUANACELIK
  12. 📚 Announced March 11, 2024 Agentic Pipelines that can: Invoke

    the right tool, pipeline/branch Self reflect Correct Iterate through tasks Haystack 2.0: DAG @TUANACELIK /IN/TUANACELIK
  13. BUILDING TIME SOME PREP Question Answering Given the documents, answer

    the question. Documents: {{ documents}} Question: {{ question }} Summarization Summarize the following text. Text: {{ text}} Question Generation Given the following document, generate some questions Document: {{ document}} Question: Translation Translate the following text to French Text: {{ text }} PROMPTING (JINJA TEMPLATING) @TUANACELIK /IN/TUANACELIK
  14. BUILDING TIME SOME PREP COMPONENTS EMBEDDERS GENERATORS PREPROCESSORS PROMPT BUILDERS

    CLASSIFIERS CONVERTERS .... @TUANACELIK /IN/TUANACELIK