DeepSeek R1

[mattifrind]

I'm wondering if the release of the DeepSeek R1 model changes the fact that open source models without finetuning are not usable for code review purposes (Qodo Docs).

Has anyone tried using this new model with the code review?

[mrT23]

I would just mention that the full Deepseek R1 model (685B MOE) is in my opinion "semi-open-source".
I don't see a regular user actually hosting this monster.
In practice, to use it yourself you will need to pay and utilize the API.

Or use smaller versions (like the distilled one)

[mattifrind]

Yes I was thinking about a distilled version like the ones Ollama offers. Sorry I wasn't clear about that.

[congziqi77]

This error occurs when I call /imporve or /describe while using deepseek as a model.

Err：litellm.APIError: APIError: DeepseekException - Expecting value: line 1 column 1 (char 0)\n

POST Request Sent from LiteLLM:
curl -X POST
https://api.deepseek.com/v1/
-H 'Authorization: *****'
-d '{'model': 'deepseek-chat',

The default deepseek address is https://api.deepseek.com/beta , I have modified the Base_URL in the code, but this problem still exists.

config :

[config]
model="deepseek/deepseek-chat"
fallback_models=["deepseek/deepseek-chat"]
git_provider="gitlab"

[IsmaelMartinez]

This has been overtaken by events. R1 is registered as deepseek/deepseek-reasoner with a 64K window and is already handled as a user-message-only, no-temperature model, so the chat-template quirks are covered. Distilled variants are registered as well, through DeepInfra and Groq, if you want something closer to self-hosted. On the wider question, the general guidance in the changing-a-model guide still holds: local and open models are fine for experimentation and /ask, but commercial models remain the recommendation for production review quality. The base-URL error reported in the last comment was against a 2025 build and is not reproducible against current DeepSeek support; please open a fresh bug with logs if you still see it. Closing as answered.