Bending_LLMs_To_Your_Will__and_your_will_is_to_create_structured_data_.pdf

Transcript

2. Tuana Çelik Hosting LLM Apps @ Scale with Haystack, Titan ML & Jina AI Bending LLMs To Your Will …(and your will is to create structured data) Open NLP Meetup
3. 01 - What is it? • Twitter: @tuanacelik • Linkedin: Tuana Celik • GitHub: @tuanacelik • 🏗 Open source LLM Framework: Haystack • 📍 Amsterdam, Netherlands Tuana Çelik Lead Developer Advocate 🥑 deepset
4. 1 Using LLMs to extract information in a structured format 2 3 Validating JSON outputs Haystack: Looping Pipelines and Custom Components
5. Getting to the key bits of info i.e. get to the point my friend
6. Getting to the key bits of info i.e. get to the point my friend
7. Getting to the key bits of info i.e. get to the point my friend
8. Getting to the key bits of info i.e. get to the point my friend
9. Getting to the key bits of info i.e. get to the point my friend
10. Validating the output Tutorial 📓
11. Validating the output JSON Schema Meeting Notes
12. Validating the output JSON Schema Meeting Notes
13. Validating the output JSON Schema Meeting Notes
14. Validating the output JSON Schema Meeting Notes Pydantic
15. Validating the output JSON Schema Meeting Notes Pydantic ⚠
16. Validating the output JSON Schema Meeting Notes Pydantic ✅
17. • Fully open-source framework built in Python for custom LLM applications Haystack (2.0-Beta)
18. • Fully open-source framework built in Python for custom LLM applications • Provides tools that developers need to build state-of-the-art NLP systems Haystack (2.0-Beta)
19. • Fully open-source framework built in Python for custom LLM applications • Provides tools that developers need to build state-of-the-art NLP systems • Building blocks: Pipelines & Components Haystack (2.0-Beta)
20. Pipelines
21. Pipelines
22. Pipelines… they can loop
23. • Haystack provides a simple interface for the creation of custom components. • These components can have more than one output, or outputs that changes depending on certain conditions Haystack (2.0-Beta): Custom Components
24. Customized Prompts Prompt Given the following context, answer the question. If the answer is not contained within the context, say ‘I don’t know’. Question Answering Context: {{context}} Question: {{question}} Answer: Prompt Summarize the following text. Summarization Text: {{text}} Summary: Question Generation Document: {{document}} Questions: Prompt Given the following document, generate some questions
25. Validated Structured Output Generator Recipe 󰟷 • A “Validator” component • A very clever prompt • An LLM • A pipeline to stitch them all
26. Custom Component: Output Validator Recipe 󰟷 A “Validator” component ✅ A very clever prompt An LLM A pipeline
27. Custom Component: Output Validator Recipe 󰟷 A “Validator” component ✅ A very clever prompt An LLM A pipeline
28. Customized Prompt: validated structured output Recipe 󰟷 A “Validator” component ✅ A very clever prompt ✅ An LLM A pipeline
29. Customized Prompt: validated structured output Recipe 󰟷 A “Validator” component ✅ A very clever prompt ✅ An LLM A pipeline
30. The LLM: a Generator Component Recipe 󰟷 A “Validator” component ✅ A very clever prompt ✅ An LLM ✅ A pipeline Generators 📓
31. The Pipeline Recipe 󰟷 A “Validator” component ✅ A very clever prompt ✅ An LLM ✅ A pipeline ✅
32. The Result Try it 👇
33. Resources Haystack 👇 Haystack 2.0 Docs 👇 Thank you! @Haystack_AI @tuanacelik /in/tuanacelik