Incorporating LLMs into practical NLP workflows

Slide 1

Slide 1 text

Ines Montani Explosion incorporating   llms into practical   nlp workflows

Slide 2

Slide 2 text

spaCy Open-source library for industrial-strength Natural Language Processing 100k+ USERS 130m+ DOWNLOADS → spacy.io

Slide 3

Slide 3 text

→ spacy.io

Slide 4

Slide 4 text

prodigy Annotation tool for creating training data for machine learning models 8000+ USERS → prodigy.ai

Slide 5

Slide 5 text

→ prodigy.ai

Slide 6

Slide 6 text

incorporating   llms* into practical   nlp workflows * large language models

Slide 7

Slide 7 text

practical   workflows

Slide 8

Slide 8 text

• supervised learning practical   workflows

Slide 9

Slide 9 text

• supervised learning • tell computers exactly what to do practical   workflows

Slide 10

Slide 10 text

• supervised learning • tell computers exactly what to do • needs enough good data practical   workflows

Slide 11

Slide 11 text

• supervised learning • tell computers exactly what to do • needs enough good data • ML + business logic practical   workflows

Slide 12

Slide 12 text

LLMs as a tool #1 specific is better

Slide 13

Slide 13 text

faster is better LLMs as a tool #2

Slide 14

Slide 14 text

private is better LLMs as a tool #3

Slide 15

Slide 15 text

better is better LLMs as a tool #4

Slide 16

Slide 16 text

problems

Slide 17

Slide 17 text

problems • prompt engineering

Slide 18

Slide 18 text

problems • prompt engineering • inconsistent results

Slide 19

Slide 19 text

problems • prompt engineering • inconsistent results • unstructured responses

Slide 20

Slide 20 text

working   with llms

Slide 21

Slide 21 text

working   with llms • iterative (prompting, parsing)

Slide 22

Slide 22 text

working   with llms • iterative (prompting, parsing) • evaluation is extremely important

Slide 23

Slide 23 text

working   with llms • iterative (prompting, parsing) • evaluation is extremely important • improve, not replace task-specific models

Slide 24

Slide 24 text

working   with llms • iterative (prompting, parsing) • evaluation is extremely important • improve, not replace task-specific models scriptable workflows

Slide 25

Slide 25 text

working   with llms • iterative (prompting, parsing) • evaluation is extremely important • improve, not replace task-specific models scriptable workflows human in the loop

Slide 26

Slide 26 text

working   with llms • iterative (prompting, parsing) • evaluation is extremely important • improve, not replace task-specific models scriptable workflows human in the loop business logic

Slide 27

Slide 27 text

→ github.com/explosion/prodigy-openai-recipes

Slide 28

Slide 28 text

→ prodigy.ai

Slide 29

Slide 29 text

→ prodigy.ai query LLM and parse response

Slide 30

Slide 30 text

→ prodigy.ai query LLM and parse response tune prompt if needed

Slide 31

Slide 31 text

→ prodigy.ai

Slide 32

Slide 32 text

→ prodigy.ai

Slide 33

Slide 33 text

→ prodigy.ai correct mistakes

Slide 34

Slide 34 text

→ prodigy.ai correct mistakes

Slide 35

Slide 35 text

→ prodigy.ai correct mistakes add correct answer to prompt to tune it

Slide 36

Slide 36 text

→ prodigy.ai

Slide 37

Slide 37 text

→ prodigy.ai generate and display reason

Slide 38

Slide 38 text

→ prodigy.ai

Slide 39

Slide 39 text

reality is not   an end-to-end   prediction problem

Slide 40

Slide 40 text

“Microsoft acquires software development platform GitHub for $7.5 billion”

Slide 41

Slide 41 text

“Microsoft acquires software development platform GitHub for $7.5 billion”

Slide 42

Slide 42 text

TEXT CLASSIFIER “Microsoft acquires software development platform GitHub for $7.5 billion”

Slide 43

Slide 43 text

TEXT CLASSIFIER ENTITY RECOGNIZER “Microsoft acquires software development platform GitHub for $7.5 billion”

Slide 44

Slide 44 text

TEXT CLASSIFIER ENTITY RECOGNIZER ENTITY LINKER “Microsoft acquires software development platform GitHub for $7.5 billion”

Slide 45

Slide 45 text

TEXT CLASSIFIER ENTITY RECOGNIZER ENTITY LINKER ATTRIBUTE LOOKUP “Microsoft acquires software development platform GitHub for $7.5 billion”

Slide 46

Slide 46 text

TEXT CLASSIFIER ENTITY RECOGNIZER ENTITY LINKER ATTRIBUTE LOOKUP CURRENCY NORMALIZER “Microsoft acquires software development platform GitHub for $7.5 billion”

Slide 47

Slide 47 text

TEXT CLASSIFIER ENTITY RECOGNIZER ENTITY LINKER ATTRIBUTE LOOKUP CURRENCY NORMALIZER “Microsoft acquires software development platform GitHub for $7.5 billion” * *

Slide 48

Slide 48 text

→ github.com/explosion/prodigy-openai-recipes summary

Slide 49

Slide 49 text

→ github.com/explosion/prodigy-openai-recipes summary • LLMs are a great tool for creating better data   faster and iteratively

Slide 50

Slide 50 text

→ github.com/explosion/prodigy-openai-recipes summary • LLMs are a great tool for creating better data   faster and iteratively • you’ll always need task-specific data

Slide 51

Slide 51 text

→ github.com/explosion/prodigy-openai-recipes summary • LLMs are a great tool for creating better data   faster and iteratively • you’ll always need task-specific data • many new applications in the future

Slide 52

Slide 52 text

future   work

Slide 53

Slide 53 text

future   work • data structures for result parsing

Slide 54

Slide 54 text

future   work • data structures for result parsing • workflows for robust evaluation

Slide 55

Slide 55 text

future   work • data structures for result parsing • workflows for robust evaluation • interactive prompt testing

Slide 56

Slide 56 text

future   work • data structures for result parsing • workflows for robust evaluation • interactive prompt testing • support for open-source models

Slide 57

Slide 57 text

💥 Explosion   explosion.ai 📲 Twitter   @_inesmontani 📲 Mastodon   @[email protected] thank you!