4 ms·

Have you tested this on big models involving multi-gpu communication, or any plans?

• ipiszy 4y ago
  For now it's for single GPU inference only.