LMQL?

Test of the features that LMQL provides: (Language Model Query Language)

Comment

Challenge 4: 🧠 Teach Ivan 3 new things about LMQL 🧠

Prize: $100 LMQL: https://lmql.ai/

Description:

The goal is to teach Ivan 3 new things about LMQL features, how it works, or just anything interesting about it under the hood. This is an EXTREMELY SUBJECTIVE one, so take that into consideration You can call ivan over once an hour and present him your findings, he will evaluate how many of them are actually interesting and new Making a DEMO gives you the best chance at winning! First team to reach 3 wins.

What it does

This was my endeavor and exposure to AI. I wanted to see what LMQL was, I thought this challenge would teach me and Ivan how LMQL works.

How we built it

Using LMQL Playground and OpenAI API.

Challenges we ran into

What is AI? What is transformations? What is a model? How to use OpenAI API?

Accomplishments that we're proud of

I learned some basics of tokenization, new technologies and how to use OpenAI API.

What we learned

I will link photos of my notes to this devpost.

  • LMQL basics
  • LLM
  • Tokenization
  • Types of token-level constraints
  • Types LLM training
  • GPT
  • Instruct
  • RHLS
  • Types of decoders (greedy vs beam)
  • beam hyperparameters
  • Sample (where it is randomized based on temperature, higher = more random)

Tokens

  • units of text that the models process & generate
  • represent individual characters and words
  • token-level operations

LMQL is meant for manipulating at the token level, so 'real' LMQL happens in local models

What's next for LMQL?

LEARN MORE ABOUT AI

Built With

Share this project:

Updates