Deep Dive Into The Toolformer

This week we cover the "Toolformer: Language Models Can Teach Themselves to Use Tools" paper from Meta and Universitat Pompeu Fabra. This paper shows how you can train your LLM to use tools like a calculator and calendar through API calls.

--

Get Oxen 🐂 https://oxen.ai/

Oxen.ai makes versioning your datasets as easy as versioning your code! Even is millions of unstructured images, we quickly handle any type of data so you can build cutting-edge AI.

--

Toolformer 📜 https://arxiv.org/abs/2302.04761

The Datasets 🔢

https://www.oxen.ai/Laurence/mlqa
https://www.oxen.ai/Laurence/lama
https://www.oxen.ai/Laurence/ASDiv
https://www.oxen.ai/Laurence/SVAMP
https://www.oxen.ai/Laurence/web_ques...
https://www.oxen.ai/Laurence/MAWPS
https://www.oxen.ai/Laurence/templama
https://www.oxen.ai/datasets/OxenAI-P...

Filtering Functions ✂️ https://github.com/lucidrains/toolfor...

Toolformer Notes 📜 https://www.oxen.ai/blog/toolformer-l...

Join Arxiv Dives 🤿 https://oxen.ai/community

Discord 🗿   / discord  

--

Chapters
0:00 Intro to the Toolformer
6:40 Toolformer Architecture
7:43 Approach
9:39 Creating the Training Data
12:24 Generate API Call Data
13:36 Together AI Demo
15:35 Axiv Paper Examples
18:00 Execute API Calls
19:53 Filtering API Calls and Math
31:15 Experiments
32:12 Results
34:14 Scaling Laws
35:22 Questions