Thread Rating:
  • 0 Vote(s) - 0 Average
  • 1
  • 2
  • 3
  • 4
  • 5
KV cache compression worth it
#11
Tried to reproduce it just now. On mini PC with PyTorch, I saw basically the same behavior. The main thing that moved was latency, not quality.
Reply
#12
I'd approach it in a boring order. Before rebuilding anything, I'd test one clean baseline with Phi and compare from there.
M3 Max • MLX • local-first
Reply


Forum Jump:


Users browsing this thread: