Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

deepseek-r1:32b or 14b almost cannot work in roo cline, because it never calls the tools leading to continuous errors #28

Open
DoiiarX opened this issue Jan 21, 2025 · 5 comments

Comments

@DoiiarX
Copy link

DoiiarX commented Jan 21, 2025

deepseek-r1:32b or 14b almost cannot work in roo cline, because it never calls the tools leading to continuous errors

@LYouC
Copy link

LYouC commented Jan 22, 2025

I also encountered this issue, which might be a problem with the model's capability. I once called qwen2-coder-32b, and it worked fine in roocline. However, deepseek-r1:32b would report an error after multiple repeated invalid conversations.

@leobarcellos
Copy link

I could not make it work even with the deepseek-r1:70b

@jp-gorman
Copy link

I have the same issue using cline and the R1-32B model locally on ollama using ollama published R1 model.

@Zodaztream
Copy link

Zodaztream commented Jan 22, 2025

For me the issue was context length, the default context length is around 4K for Ollama. You can fix this by creating a Modelfile with the following:

FROM deepseek-r1:14b
PARAMETER num_ctx 32768

and then create a model from this with a unique name using ollama. Choose this model in cline and it should start working.

You can also use LM studio which allows to increase context length by holding alt over the model before loading it.

These two approaches will make it work.

@jwadow
Copy link

jwadow commented Jan 22, 2025

For me the issue was context length, the default context length is around 4K for Ollama. You can fix this by creating a Modelfile with the following:

FROM deepseek-r1:14b
PARAMETER num_ctx 32768

and then create a model from this with a unique name using ollama. Choose this model in cline and it should start working.

Yes, thank you, it worked and became much better. For 14B I set num_ctx 25000 and it is able to complete the simple task, albeit after a long time.
7B and especially 1.5B with 32768 are not able to complete the simple task.

R1 models seem to not understand that they need to work with an open active file and replace 1 variable there (in "Code" mode).

But there are still certain technical shortcomings in roo cline:

  1. The model constantly uses visible tags like <tool_use> and <ask_ollowup_question>.
  2. For some reason, the beginning of the answer is very long. I think the problem is that this is a <think> block that is invisible to the user? It would be nice to add a pop-up inscription "Thinking", if so.

Oh, I noticed that this is not a repo of roo cline, but of the creator of the model.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

6 participants