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/391 - Fix DeepSeek q8_0 cache.md
Thomas eaa2510a28 Add GitHub data: filename sanitization (#640)
2025-07-23 13:31:53 +02:00

606 B
Raw Permalink Blame History

🐛 #391 - Fix DeepSeek q8_0 cache

Author ikawrakow
State ❌ Closed
Created 2025-05-07
Updated 2025-05-07

Description

Nobody has used ik_llama.cpp with a DeepSeek model and Q8_0 KV cache since PR #351?

This PR fixes the assert one gets when one tries to use a DeepSeek model on the CPU using Q8_0 KV cache.

Also, it seems the optimization I added in #351 to repack the K cache to Q8_0_R8 seems to lower TG performance for DeepSeek models, so disabling it.