Starling-7B: UC Berkeley's Open-Source LLM trained by RLAIF

In this tutorial video, I've put Starling-7B LLM to the test, showcasing its capabilities in various applications, from question answering to coding tasks. Starling-7B is a remarkable open large language model developed through Reinforcement Learning from AI Feedback (RLAIF). This model leverages the cutting-edge GPT-4 labelled ranking dataset called Nectar, combined with a novel reward training and policy tuning pipeline.
One of the most impressive feats of Starling-7B-alpha is its outstanding performance, scoring 8.09 in the MT Bench evaluation with GPT-4 as the judge. It outshines every other model in the MT-Bench except for OpenAI's GPT-4 and GPT-4 Turbo, making it a significant advancement in the field of natural language understanding.
I'm excited to share this video with you, where you'll witness Starling-7B's capabilities first-hand.
Don't forget to like, comment with your thoughts, and subscribe for more insightful content. Stay tuned for future updates and tutorials, and explore the power of RLAIF-driven LLMs!
Notebook: github.com/AIA...
Starling 7B HF: huggingface.co...
Starling Website: starling.cs.be...
#generativeai #ai #python

[@user4-j1w]

Can I use this through langchain?

[@sjimosui8279]

Would love to see this starling model with RAG,
I recently came across Dragon models by llmware which are models finetuned on RAG.
Could you please check them out?

[@AIAnytime]

Yes working on it.

[@khalidal-reemi3361]

Great Videos as usual.
Which is better for RAG? Falcon 7B or Starling ?

[@sneharoy3566]

Fantastic

[@AIAnytime]

Thank you! Cheers!