github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 08:30:05 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #2495] Llama2: q4_km as default? #1457

New Issue
Closed
opened 2026-04-12 11:21:14 -05:00 by GiteaMirror · 5 comments
GiteaMirror commented 2026-04-12 11:21:14 -05:00
Owner
Copy Link

Originally created by @matthiasgeihs on GitHub (Feb 14, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/2495

Just saw this table: https://github.com/ggerganov/llama.cpp/pull/1684#issuecomment-1579252501

Perplexity loss is considerably higher for q4_0 compared to q4_km. km provides the best tradeoff between size and performance, as noted also here.

ollama currently uses q4_0 as the default for llama2: https://ollama.com/library/llama2:latest (as observed by comparing the ID 78e26419b446).

Suggestion:
Use q4_km as the default instead.

Originally created by @matthiasgeihs on GitHub (Feb 14, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/2495 Just saw this table: https://github.com/ggerganov/llama.cpp/pull/1684#issuecomment-1579252501 Perplexity loss is considerably higher for `q4_0` compared to `q4_km`. `km` provides the best tradeoff between size and performance, as noted also [here](https://huggingface.co/TheBloke/Mixtral-8x7B-v0.1-GGUF). `ollama` currently uses `q4_0` as the default for `llama2`: https://ollama.com/library/llama2:latest (as observed by comparing the ID 78e26419b446). **Suggestion:** Use [`q4_km`](https://ollama.com/library/llama2:7b-chat-q4_K_M) as the default instead.
GiteaMirror added the feature request label 2026-04-12 11:21:14 -05:00
GiteaMirror commented 2026-04-12 11:21:15 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (May 17, 2024):

Hey @matthiasgeihs you can pull 7b-chat-q4_K_M which is the model you want. The ollama cp command will allow you to change it on your own host to the default. We've been debating internally for a while about changing the default, but we'll see.

I'll go ahead and close the issue though.

<!-- gh-comment-id:2116460497 --> @pdevine commented on GitHub (May 17, 2024): Hey @matthiasgeihs you can pull [7b-chat-q4_K_M](https://ollama.com/library/llama2:7b-chat-q4_K_M) which is the model you want. The `ollama cp` command will allow you to change it on your own host to the default. We've been debating internally for a while about changing the default, but we'll see. I'll go ahead and close the issue though.
GiteaMirror commented 2026-04-12 11:21:15 -05:00
Author
Owner
Copy Link

@matthiasgeihs commented on GitHub (May 17, 2024):

Thx for the update. An option to change the default quantization on my machine would be great. Are you saying ollama cp can do that? How?

<!-- gh-comment-id:2116484865 --> @matthiasgeihs commented on GitHub (May 17, 2024): Thx for the update. An option to change the default quantization on my machine would be great. Are you saying `ollama cp` can do that? How?
GiteaMirror commented 2026-04-12 11:21:15 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (May 19, 2024):

Ultimately I'd love you to be able to just set a default and not have to worry about tags any more (I demo'd something like this at the Ollama + Friends meetup in Paris last month). For now if you don't want to type out the whole tag name, just ollama cp llama2:7b-chat-q4_K_M mymodel which would allow you to ollama run mymodel as a convenient shortcut.

<!-- gh-comment-id:2119098752 --> @pdevine commented on GitHub (May 19, 2024): Ultimately I'd love you to be able to just set a default and not have to worry about tags any more (I demo'd something like this at the Ollama + Friends meetup in Paris last month). For now if you don't want to type out the whole tag name, just `ollama cp llama2:7b-chat-q4_K_M mymodel` which would allow you to `ollama run mymodel` as a convenient shortcut.
GiteaMirror commented 2026-04-12 11:21:16 -05:00
Author
Owner
Copy Link

@ncoquelet commented on GitHub (Jun 7, 2024):

👍 Thanks for the tips on the ollama cp command. but that double our models list, not really convincing with many models
🤞 For an option to customise the default quantization tag for all models at once when pull and run

<!-- gh-comment-id:2154369308 --> @ncoquelet commented on GitHub (Jun 7, 2024): :+1: Thanks for the tips on the `ollama cp` command. but that double our models list, not really convincing with many models :crossed_fingers: For an option to customise the default quantization tag for all models at once when pull and run
GiteaMirror commented 2026-04-12 11:21:16 -05:00
Author
Owner
Copy Link

@ncoquelet commented on GitHub (Jun 7, 2024):

linked open issue https://github.com/ollama/ollama/issues/1543

<!-- gh-comment-id:2154382739 --> @ncoquelet commented on GitHub (Jun 7, 2024): linked open issue https://github.com/ollama/ollama/issues/1543
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 feature request
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#1457
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?