support deepseek / reasoning grade models

[webcoderz]

ive been trying to brainstorm how to support reasoning grade models here, ive tried it with deepseek and the deepseek distils and it cant seem to mark task complete and stop the orchestration like llama etc can. thoughts?

[zzstoatzz]

hi @webcoderz - can you be more specific on what the request is? we don't implement model support ourselves, we use pydantic-ai.

you can use deepseek today

https://github.com/PrefectHQ/marvin/blob/main/examples/deepseek_chat.py

if you're running into a specific issue and you have an MRE I'd be happy to take a look

[webcoderz]

Oh I’m trying to build an agent with Deepseek but it doesn’t seem able to stop the orchestrator and mark the task successfully so it just infinite loops until I stop it. Wasn’t trying to do interactive tasks, moreso singular discrete tasks until successful

[zzstoatzz]

oh okay, do you have a specific example that you were trying?

[webcoderz]

Ehh I was just doing some generic task to test ; it wasn’t anything crazy, like: read this task and summarize, or something similar. It got into this infinite reasoning loop about accomplishing the task and it should end the task.

[zzstoatzz]

oh ok i just meant a specific example because then I could run it and potentially reproduce the issues you were seeing

[webcoderz]

Honestly I don’t remember anymore it was just some generic task I gave it to check it out..