github-starred/ollama
2
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-05-06 16:11:34 -05:00
Code Issues 14.5k Packages Projects Releases 120 Wiki Activity

[GH-ISSUE #1919] create model, not meeting the performance requirements of the gguf #63141

New Issue
Closed
opened 2026-05-03 12:16:59 -05:00 by GiteaMirror · 2 comments
GiteaMirror commented 2026-05-03 12:16:59 -05:00
Owner
Copy Link

Originally created by @quanpinjie on GitHub (Jan 11, 2024).
Original GitHub issue: https://github.com/ollama/ollama/issues/1919

i convert baichuan2 to gguf and create a model,

The result is poor performance，do I need to configure anything else

modelfile:

FROM ./baichuan2-ggml-model-f16.gguf

image

Originally created by @quanpinjie on GitHub (Jan 11, 2024). Original GitHub issue: https://github.com/ollama/ollama/issues/1919 i convert baichuan2 to gguf and create a model, The result is poor performance，do I need to configure anything else modelfile: FROM ./baichuan2-ggml-model-f16.gguf ![image](https://github.com/jmorganca/ollama/assets/2564119/ea70b5b6-9729-4a93-b990-a4ce439e6921)
GiteaMirror commented 2026-05-03 12:16:59 -05:00
Author
Owner
Copy Link

@easp commented on GitHub (Jan 11, 2024):

Poor performance compared to what? What hardware are you running on?

It looks like you are using the fp16 version of the model. That will require a lot of VRAM and memory bandwidth. Try a q4_k_m quantization.

<!-- gh-comment-id:1887976708 --> @easp commented on GitHub (Jan 11, 2024): Poor performance compared to what? What hardware are you running on? It looks like you are using the fp16 version of the model. That will require a lot of VRAM and memory bandwidth. Try a q4_k_m quantization.
GiteaMirror commented 2026-05-03 12:17:00 -05:00
Author
Owner
Copy Link

@pdevine commented on GitHub (Mar 12, 2024):

Hey @quanpinjie, sorry for taking so long to comment on the issue. The reason why the performance is so slow is because you need to "quantize" the model for better performance. We created a document here which explains the steps you need to take. By default Ollama uses q4_0 as the default quantization level, so if you want to see similar performance you should try using that level.

I'm going to go ahead and close out the issue, but please feel free to keep commenting or reopen it if I didn't answer your question.

<!-- gh-comment-id:1992691498 --> @pdevine commented on GitHub (Mar 12, 2024): Hey @quanpinjie, sorry for taking so long to comment on the issue. The reason why the performance is so slow is because you need to "quantize" the model for better performance. We created a document [here](https://github.com/ollama/ollama/blob/main/docs/import.md#quantize-the-model) which explains the steps you need to take. By default Ollama uses `q4_0` as the default quantization level, so if you want to see similar performance you should try using that level. I'm going to go ahead and close out the issue, but please feel free to keep commenting or reopen it if I didn't answer your question.
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
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#63141
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?