Same model, same Q4_K_M label: 5.02, 5.07 and 5.27 bits per weight

Y	Hacker News new \| ask \| show \| jobs

	Same model, same Q4_K_M label: 5.02, 5.07 and 5.27 bits per weight (github.com)
	19 points by logickkk1 15 days ago

1 comments

p_stuart82 14 days ago

Does this catch intermittent fallback between requests?

link

logickkk1 14 days ago

Yes, picchio monitors a running llama-server on a timer and flags when the prefill/decode ratio goes CPU-shaped. I hot-swapped one mid-run. probe 4 caught ENGAGED -> NOT ENGAGED.

link