← back

Building State of the Art Open Weights Tool Use: The Command R Family: Sandra Kublik

1.9K views · Aug 26, 2024 · 15:03 min · Watch on YouTube ↗
Takeaway

Command R+ proved open-weight models can match frontier closed models on tool use and grounded RAG, with structured citation support.

Summary

  • Cohere's Command R (Mar 11) and Command R+ are open-weight models optimized for RAG and tool use; competitive with GPT-4 Turbo and Claude Opus while small enough to run locally.
  • Command R+ trended on OpenRouter within 2 weeks; ~150k HuggingFace downloads; became the HuggingChat backbone with parser, image editor, calculator tools.
  • Design choices: prompt-sensitive grounding for RAG (model must distinguish conversation history from retrieved info), structured tool-use citations, multilingual support.
  • ~500k developers/researchers using the R family at talk time.
open-weightstool-userag
Original description
We opened model weights for Command R and R+, and the response was incredible. This talk will showcase the community's innovative projects and our journey to building SOTA multi-step tool use proficiency with the R family. We will also share the design decisions that make the R family unique and effective.

Recorded live in San Francisco at the AI Engineer World's Fair. See the full schedule of talks at https://www.ai.engineer/worldsfair/2024/schedule & join us at the AI Engineer World's Fair in 2025! Get your tickets today at https://ai.engineer/2025

About Sandra
Sandra Kublik is a member of the Developer Relations team at Cohere, where she educates and leads the LLM community. In 2023, Sandra co-authored the first manual book on OpenAI API, "GPT-3: The Ultimate Guide to Building NLP Products with OpenAI API." She hosts a YouTube channel and podcast where she explores the latest advancements and applications of LLMs.