ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-01-26 09:09:50 +00:00
Code Issues Packages Projects Releases Wiki Activity
Files
main
ik_llama.cpp/github-data/pull_requests/20 - iq2_k_ slightly better bpw - accuracy compromise.md
Thomas eaa2510a28 Add GitHub data: filename sanitization (#640)
2025-07-23 13:31:53 +02:00

506 B
Raw Permalink Blame History

🔀 #20 - iq2_k: slightly better bpw - accuracy compromise

Author ikawrakow
State ❌ Closed
Created 2024-08-19
Updated 2024-08-19

Description

For LLaMA-3.1 models:

  • It is better to quantize all of attn_v with iq3_k instead of half of attn_v with iq4_k
  • Quantizing attn_output with iq3_k results in a larger PPL decrease compared to what one expects from the added bpw.