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

feedback: mmap for keeping Model in VRAM when Flash Attention is used #1717

Open
dan-homebrew opened this issue Nov 23, 2024 · 1 comment
Open

feedback: mmap for keeping Model in VRAM when Flash Attention is used #1717

dan-homebrew opened this issue Nov 23, 2024 · 1 comment
Assignees
@vansangpfiev
Labels
engine: llama.cpp

Comments

@dan-homebrew
Copy link
Contributor

dan-homebrew commented Nov 23, 2024 •
edited
Loading

Goal

Feedback from WiseFarAI:

I am also wondering if you guys use Mmap and other ways of keeping the model in Vram memory and if/when Flash Attention is used. Since these parameters cannot easily be observed (their current setting) or changed from within the UI. Sometimes it seems like it tries to reload the model mid-conversation and the generation speed drops from 6 tokens to 2 per second.
@dan-homebrew dan-homebrew converted this from a draft issue Nov 23, 2024
@dan-homebrew dan-homebrew changed the title feedback: mmap for keeping Model in VRAM feedback: mmap for keeping Model in VRAM when Flash Attention is used Nov 23, 2024
@gabrielle-ong gabrielle-ong mentioned this issue Nov 28, 2024
Closed
@gabrielle-ong
Copy link
Contributor

@vansangpfiev assigning to you to investigate, cc @dan-homebrew - will have a call with Sang next week

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@vansangpfiev vansangpfiev

Labels
engine: llama.cpp
Projects
Jan & Cortex
Status: Investigating
Milestone
No milestone
Development

No branches or pull requests
3 participants
@vansangpfiev @gabrielle-ong @dan-homebrew