vicgalle/zero-shot-reward-models

ZYN: Zero-Shot Reward Models with Yes-No Questions

40
/ 100
Emerging

This project helps developers fine-tune large language models (LLMs) to generate specific types of text, like positive movie reviews or non-toxic content. You provide a prompt that asks a yes/no question about the desired text quality, and the system trains an LLM to produce outputs that consistently elicit a 'yes' answer. This is for machine learning engineers and researchers working on text generation applications.

No commits in the last 6 months.

Use this if you need to guide an instruction-tuned LLM to generate text that fulfills a specific, measurable quality or attribute without needing to collect extensive human preference data for a reward model.

Not ideal if you're not a developer working with large language models, or if your desired text quality is highly subjective and cannot be framed as a clear yes/no question.

LLM fine-tuning text generation reinforcement learning content moderation sentiment control
Stale 6m No Package No Dependents
Maintenance 0 / 25
Adoption 7 / 25
Maturity 16 / 25
Community 17 / 25

How are scores calculated?

Stars

35

Forks

8

Language

Python

License

MIT

Last pushed

Aug 15, 2023

Commits (30d)

0

Get this data via API

curl "https://pt-edge.onrender.com/api/v1/quality/transformers/vicgalle/zero-shot-reward-models"

Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.