Comment by numpad0

4 days ago

Even in tensor parallel modes? I thought it could only work if you're fine stalling all but n GPU for n users at any given moments.

0 comments