Urgent.News

What's breaking now, across thousands of outlets.

More in AI

An 80GB H100, and only 3B of the model could keep up with conversation

📝 Originally published (in Japanese) at forge.workstyle.tech . I used to think that 80 GB of VRAM would be enough to chat with a large model.

  • An 80GB H100 GPU insufficient for full 35B model
  • 3B-parameter model used for each token generation
  • MoE mechanism enabled 35B model with 3B weights per token

More from Monday 12 October →