github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-07 00:22:43 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #4626] about model quantization #2905

New Issue
Closed
opened 2026-04-12 13:15:34 -05:00 by GiteaMirror · 3 comments
GiteaMirror commented 2026-04-12 13:15:34 -05:00
Owner
Copy Link

Originally created by @andyyumiao on GitHub (May 25, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/4626

What are the quantization parameters used for the llama3 model in Ollama? For example, llama3 version, quantization parameters, etc
The llama3 8b version that I quantified using llama.cpp myself is not as good as the llama3 8b version that comes with Ollama, so I want to know the reason.

Originally created by @andyyumiao on GitHub (May 25, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/4626 What are the quantization parameters used for the llama3 model in Ollama? For example, llama3 version, quantization parameters, etc The llama3 8b version that I quantified using llama.cpp myself is not as good as the llama3 8b version that comes with Ollama, so I want to know the reason.
GiteaMirror added the question label 2026-04-12 13:15:34 -05:00
GiteaMirror commented 2026-04-12 13:15:35 -05:00
Author
Owner
Copy Link

@umiyuki commented on GitHub (May 25, 2024):

Conversely, Phi-3-medium is degraded in the ollama version. it has lower benchmark scores than Llama.cpp. several people have tried Phi-3-medium on ollama and found the performance to be poor. Why is this happening? Either the inference parameter settings or the chat template is wrong or something is broken in quantization.

<!-- gh-comment-id:2130726686 --> @umiyuki commented on GitHub (May 25, 2024): Conversely, Phi-3-medium is degraded in the ollama version. it has lower benchmark scores than Llama.cpp. several people have tried Phi-3-medium on ollama and found the performance to be poor. Why is this happening? Either the inference parameter settings or the chat template is wrong or something is broken in quantization.
GiteaMirror commented 2026-04-12 13:15:36 -05:00
Author
Owner
Copy Link

@skye0402 commented on GitHub (May 25, 2024):

@umiyuki I found the 4K 14b model to be working quite well, the 128k 14b model definitely has problems. I used 6 bit quantization for both to compare with identical input.

<!-- gh-comment-id:2131227446 --> @skye0402 commented on GitHub (May 25, 2024): @umiyuki I found the 4K 14b model to be working quite well, the 128k 14b model definitely has problems. I used 6 bit quantization for both to compare with identical input.
GiteaMirror commented 2026-04-12 13:15:37 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (May 28, 2024):

@andyyumiao the default is Q4_0, which you can see here:
Screenshot 2024-05-28 at 13 33 02
Screenshot 2024-05-28 at 13 31 27

It's possible (and more likely) that your prompt template is off on the version that you converted yourself. I'll close out the issue.

<!-- gh-comment-id:2136068708 --> @pdevine commented on GitHub (May 28, 2024): @andyyumiao the default is `Q4_0`, which you can see here: <img width="776" alt="Screenshot 2024-05-28 at 13 33 02" src="https://github.com/ollama/ollama/assets/75239/a5c09e23-f853-4600-9326-4b62be121c21"> <img width="777" alt="Screenshot 2024-05-28 at 13 31 27" src="https://github.com/ollama/ollama/assets/75239/5fc036a5-c844-4dad-89ff-de5c1990ff58"> It's possible (and more likely) that your prompt template is off on the version that you converted yourself. I'll close out the issue.
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 question
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#2905
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?