Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Use Q4_K for attn_v for Q2_K_S when n_gqa >= 4 #4996

Merged
merged 1 commit into from
Jan 17, 2024
Merged

Use Q4_K for attn_v for Q2_K_S when n_gqa >= 4 #4996

merged 1 commit into from
Jan 17, 2024
+6 −1

Conversation

ikawrakow
Copy link
Contributor

I have missed this tweak when adding Q2_K_S.

With this change, model size for Mistral-7B increases by only ~30 MB (0.03 bpw) while

  • Perplexity for a context of 512 on wiki.test.raw goes down from 6.9259 to 6.7116
  • 10-shot HellaSwag score after 2000 tasks increases by 0.95 +/- 0.42.

@ggerganov ggerganov merged commit 2b3a665 into master Jan 17, 2024
41 of 46 checks passed
jordankanter pushed a commit to jordankanter/llama.cpp that referenced this pull request Feb 3, 2024
@ikawrakow @Kawrakow @jordankanter
llama : use Q4_K for attn_v for Q2_K_S when n_gqa >= 4 (ggerganov#4996)
ec0b035 
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
hodlen pushed a commit to hodlen/llama.cpp that referenced this pull request Apr 1, 2024
@ikawrakow @Kawrakow @hodlen
llama : use Q4_K for attn_v for Q2_K_S when n_gqa >= 4 (ggerganov#4996)
d13fd96 
Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@ggerganov ggerganov ggerganov approved these changes

Assignees
No one assigned
Labels
None yet
Projects
ggml : sub-4-bit integer quantization
Status: No status
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
3 participants
@ikawrakow @ggerganov @Kawrakow