Much faster CPU prompt processing (part 2) #533
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
added 9 commits
June 17, 2025 08:24
...and that's why PPL was so high. It is also high on main. This fixes it.
|
Thanks, this is huge. I feel like this will make ~70B dense models much better for hybrid inferencing on home rigs. Hope to try some quants soon! Also holy cow the |
|
Impressive work all around! |
|
Very impressive, @ikawrakow! |
Nexesenex
pushed a commit
to Nexesenex/ik_llama.cpp.nxs
that referenced
this pull request
Jun 18, 2025
* iq4_ks 203 t/s -> 357 t/s. iq4_ks_r4 is 242 t/s. * iq4_k 175 t/s -> 353 t/s. iq4_k_r4 is 208 t/s. PPL is actually lower! * iq5_ks 180 t/s -> 359 t/s. iq5_ks_r4 is 210 t/s. PPL is actually lower - 7.4160 vs 7.4494 for LlaMA-3.1-8B-Instruct * iq5_k - accuracy loss is too big * iq5_k - there was a bug with the shifts ...and that's why PPL was so high. It is also high on main. This fixes it. * iq6_k 148 t/s -> 350 t/s. There is no iq6_k_r4 PPL is actually lower because we have a bug in the existing implementation! * iq3_k 169 t/s -> 363 t/s. iq3_k_r4 is at 200 t/s. * iq2_k 190 t/s -> 364 t/s. iq2_k_r4 is at 232 t/s. * iq2_ks 200 t/s -> 367 t/s. There is no iq2_ks_r4. Co-Authored-By: Iwan Kawrakow <[email protected]>
This was referenced Jun 21, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR is a follow up of #531 and applies the technique to
IQKquants.Here is a PP-512 performance comparison for LlaMA-3.1-8B-Instruct on a Ryzen-7950X CPU between the main branch and this PR:
To put things into perspective, the fastest mainline
llama.cppquant on this CPU isQ4_0, and I get 170 t/s with today's build (build: 860a9e4ee (5688)).For a bit of history, when PR 6414 was added to
llama.cpp, it received 92 👍, 32 🎉, 34 ❤️, and 30 🚀. It only supportedQ4_0andQ8_0, and speedup compared to the master branch at the time was in the range of 40-50%, for a PP-512 of 135 t/s on the Ryzen-7950X CPU used for the above table. There was a blog post received with great fanfare on HN.