Video: AI Forum 2023 | The Small Models Revolution

[konabuta]

AI Forum 2023 | The Small Models Revolution

Link: AI Forum 2023 | The Small Models Revolution

[konabuta]

• phi-1

  • 1.3B params
  • Coding

• phi-1.5

  • 1.3B params
  • Natural Language model with NL

[konabuta]

phi-1.5 vs Falcon-7B vs Llama 2-7B

[konabuta]

Getting high quality = high educational value by filtering. Filtering by labeling small size of data and developing small classifier. And generated 1B of pure education content.

[konabuta]

Compares with major OSS LLMs. Shows near performance of 13B params LLM.