github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 08:02:14 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #1827] Massive slowdown on v 0.1.18 vs 0.1.17 with same model on Intel Mac #1041

New Issue
Closed
opened 2026-04-12 10:46:33 -05:00 by GiteaMirror · 8 comments
GiteaMirror commented 2026-04-12 10:46:33 -05:00
Owner
Copy Link

Originally created by @pjv on GitHub (Jan 6, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/1827

SCR-20240106-kfri

I don’t have exact timings but the same model (in this case, deepseek-coder:6.7b-instruct-q4_K_S) generates tokens roughly 5 times faster on 0.1.17 than on 0.1.18 on my Intel Mac.

I upgraded to 0.1.18 and noticed the slowdown in token generation and then downgraded back to 0.1.17 and immediately saw the faster throughput I am accustomed to.

Originally created by @pjv on GitHub (Jan 6, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/1827 <img width="255" alt="SCR-20240106-kfri" src="https://github.com/jmorganca/ollama/assets/327716/c1b5ce5b-acd5-4c81-a59a-6db39ff6a257"> I don’t have exact timings but the same model (in this case, `deepseek-coder:6.7b-instruct-q4_K_S`) generates tokens roughly 5 times faster on 0.1.17 than on 0.1.18 on my Intel Mac. I upgraded to 0.1.18 and noticed the slowdown in token generation and then downgraded back to 0.1.17 and immediately saw the faster throughput I am accustomed to.
GiteaMirror added the bug label 2026-04-12 10:46:33 -05:00
GiteaMirror commented 2026-04-12 10:46:34 -05:00
Author
Owner
Copy Link

@jmorganca commented on GitHub (Jan 6, 2024):

Sorry you hit this slowdown. Would it be possible to share the logs? They should be in ~/.ollama/logs/server.log - thanks so much!

<!-- gh-comment-id:1879761414 --> @jmorganca commented on GitHub (Jan 6, 2024): Sorry you hit this slowdown. Would it be possible to share the logs? They should be in `~/.ollama/logs/server.log` - thanks so much!
GiteaMirror commented 2026-04-12 10:46:35 -05:00
Author
Owner
Copy Link

@jmorganca commented on GitHub (Jan 6, 2024):

Also would it be possible to test llama2 and see if you see the same slowdown with that model architecture? Thanks!

<!-- gh-comment-id:1879762977 --> @jmorganca commented on GitHub (Jan 6, 2024): Also would it be possible to test `llama2` and see if you see the same slowdown with that model architecture? Thanks!
GiteaMirror commented 2026-04-12 10:46:35 -05:00
Author
Owner
Copy Link

@jmorganca commented on GitHub (Jan 6, 2024):

Ok! Update: I'm able to reproduce this for models with k-quants (e.g. q4_K_S, but not for regular quantization – e.g. q4_0). Will look into this!

<!-- gh-comment-id:1879765642 --> @jmorganca commented on GitHub (Jan 6, 2024): Ok! Update: I'm able to reproduce this for models with k-quants (e.g. `q4_K_S`, but not for regular quantization – e.g. `q4_0`). Will look into this!
GiteaMirror commented 2026-04-12 10:46:35 -05:00
Author
Owner
Copy Link

@pjv commented on GitHub (Jan 6, 2024):

wow, you’re a lot faster than me. I’m still generating logs for you. Do you still need them?

<!-- gh-comment-id:1879765867 --> @pjv commented on GitHub (Jan 6, 2024): wow, you’re a lot faster than me. I’m still generating logs for you. Do you still need them?
GiteaMirror commented 2026-04-12 10:46:36 -05:00
Author
Owner
Copy Link

@pjv commented on GitHub (Jan 6, 2024):

Ok! Update: I'm able to reproduce this for models with k-quants (e.g. q4_K_S, but not for regular quantization – e.g. q4_0). Will look into this!

Yup, testing the llama2 model, 0.1.18 seems a bit faster than 0.1.17. but the q4_K_S model is very slow.

<!-- gh-comment-id:1879767723 --> @pjv commented on GitHub (Jan 6, 2024): > Ok! Update: I'm able to reproduce this for models with k-quants (e.g. q4_K_S, but not for regular quantization – e.g. q4_0). Will look into this! Yup, testing the `llama2` model, 0.1.18 seems a bit faster than 0.1.17. but the q4_K_S model is very slow.
GiteaMirror commented 2026-04-12 10:46:37 -05:00
Author
Owner
Copy Link

@jmorganca commented on GitHub (Jan 6, 2024):

No worries about the logs – I can reproduce on my side. Tracking this down

<!-- gh-comment-id:1879782247 --> @jmorganca commented on GitHub (Jan 6, 2024): No worries about the logs – I can reproduce on my side. Tracking this down
GiteaMirror commented 2026-04-12 10:46:38 -05:00
Author
Owner
Copy Link

@coder543 commented on GitHub (Jan 6, 2024):

Semi-related, but isn't k-quant the newer/better quantization method? I have found it confusing that ollama defaults to the non-K quants, but maybe I'm confused about which method is better.

<!-- gh-comment-id:1879782769 --> @coder543 commented on GitHub (Jan 6, 2024): Semi-related, but isn't k-quant the newer/better quantization method? I have found it confusing that ollama defaults to the non-K quants, but maybe I'm confused about which method is better.
GiteaMirror commented 2026-04-12 10:46:39 -05:00
Author
Owner
Copy Link

@oldgithubman commented on GitHub (Apr 15, 2024):

Semi-related, but isn't k-quant the newer/better quantization method? I have found it confusing that ollama defaults to the non-K quants, but maybe I'm confused about which method is better.

Would also appreciate some insight into this

<!-- gh-comment-id:2054268428 --> @oldgithubman commented on GitHub (Apr 15, 2024): > Semi-related, but isn't k-quant the newer/better quantization method? I have found it confusing that ollama defaults to the non-K quants, but maybe I'm confused about which method is better. Would also appreciate some insight into this
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
amd
api
app
bug
build
cli
cloud
compatibility
context-length
create
docker
documentation
embeddings
feature request
feedback wanted
good first issue
gpt-oss
gpu
harmony
help wanted
image
install
intel
js
launch
linux
macos
memory
mlx
model
needs more info
networking
nvidia
ollama.com
performance
pull-request
Mirrored from GitHub Pull Request
 python
question
registry
rendering
thinking
tools
top
vulkan
windows
wsl
No Label bug
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
GiteaMirror ninjasurge
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: github-starred/ollama#1041
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?