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 #14266] Model request: Nanbeige4.1-3B #35049

New Issue
Open
opened 2026-04-22 19:13:42 -05:00 by GiteaMirror · 4 comments
GiteaMirror commented 2026-04-22 19:13:42 -05:00
Owner
Copy Link

Originally created by @section1 on GitHub (Feb 15, 2026).
Original GitHub issue: https://github.com/ollama/ollama/issues/14266

Looks good for the size.. and run it with low hardware resources.

https://huggingface.co/Nanbeige/Nanbeige4.1-3B

Thanks

Originally created by @section1 on GitHub (Feb 15, 2026). Original GitHub issue: https://github.com/ollama/ollama/issues/14266 Looks good for the size.. and run it with low hardware resources. https://huggingface.co/Nanbeige/Nanbeige4.1-3B Thanks
GiteaMirror added the model label 2026-04-22 19:13:42 -05:00
GiteaMirror commented 2026-04-22 19:13:56 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Feb 15, 2026):

This is llama architecture and so should be importable.

<!-- gh-comment-id:3904219577 --> @rick-github commented on GitHub (Feb 15, 2026): This is llama architecture and so should be [importable](https://github.com/ollama/ollama/blob/main/docs/import.mdx#Importing-a-model-from-Safetensors-weights).
GiteaMirror commented 2026-04-22 19:14:00 -05:00
Author
Owner
Copy Link

@SurealCereal commented on GitHub (Feb 15, 2026):

I tried Q6_K and Q8_0 quants on an RTX 3080 12GB with v0.16.2-rc0 and it was very slow compared to larger dense models, partly because the "thinking" traces were extremely long and strange. It argues with itself a lot for a basic prompts like "Hello. Please tell me about yourself.". I tried the recommended parameters:

temperature 0.6
top_p 0.95
repeat_penalty 1.0

I also tried top_k 0, 40 and default.

<!-- gh-comment-id:3904298527 --> @SurealCereal commented on GitHub (Feb 15, 2026): I tried [Q6_K](https://huggingface.co/mradermacher/Nanbeige4.1-3B-GGUF?show_file_info=Nanbeige4.1-3B.Q6_K.gguf) and [Q8_0](https://huggingface.co/DevQuasar/Nanbeige.Nanbeige4.1-3B-GGUF?show_file_info=Nanbeige.Nanbeige4.1-3B.Q8_0.gguf) quants on an RTX 3080 12GB with `v0.16.2-rc0` and it was very slow compared to larger dense models, partly because the "thinking" traces were extremely long and strange. It argues with itself a lot for a basic prompts like "Hello. Please tell me about yourself.". I tried the recommended parameters: ``` temperature 0.6 top_p 0.95 repeat_penalty 1.0 ``` I also tried `top_k 0, 40 and default`.
GiteaMirror commented 2026-04-22 19:14:03 -05:00
Author
Owner
Copy Link

@rick-github commented on GitHub (Feb 15, 2026):

It's only a 3b model, I would take the claims given with a pretty big piece of salt. The template in the mradermacher quant doesn't fully match the one from the original model so it's possible the response could be improved, but again, only 3b.

<!-- gh-comment-id:3904321075 --> @rick-github commented on GitHub (Feb 15, 2026): It's only a 3b model, I would take the claims given with a pretty big piece of salt. The template in the mradermacher quant doesn't fully match the one from the original model so it's possible the response could be improved, but again, only 3b.
GiteaMirror commented 2026-04-22 19:14:04 -05:00
Author
Owner
Copy Link

@section1 commented on GitHub (Feb 15, 2026):

Yeah is what i read...that is slow or very verbose when thinking.. this will improved in version 4.2 and better agentic "skills".. there's a post in reddit about this.i haven't tested for now.. but thanks for the links guys i will try to import it first and try it

<!-- gh-comment-id:3904374964 --> @section1 commented on GitHub (Feb 15, 2026): Yeah is what i read...that is slow or very verbose when thinking.. this will improved in version 4.2 and better agentic "skills".. there's a post in reddit about this.i haven't tested for now.. but thanks for the links guys i will try to import it first and try it
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 model
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#35049
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?