DeepSeek-R1-0528-Qwen3-8B optimal settings?

Posted by pigeon57434@reddit | LocalLLaMA | View on Reddit | 6 comments

Does anyone know the optimal settings for this model I'm not sure how sensitive it is I know Qwens last couple of reasoning models have been very sensitive to settings, and this is based on Qwen so

Reply to Post

6 Comments

[-]

Afraid-Employer-9331@reddit

is this model better than gemma 3 12b? someone test it please!

[-]

From Unsloth: * Set the temperature between **0.5–0.7 (0.6 recommended)** to reduce repetition and incoherence. * Set Top\_P value of **0.95 (recommended)** * For complete detailed instructions, see our guide: [Guide](https://unsloth.ai/blog/deepseek-r1-0528)

[-]

ab2377@reddit

hey thanks, i just tried this model for the first time, it doesnt support /no think?

[-]

Admirable_Bag8004@reddit

Just tried myself. Looks like you're right, I can't force the model not to use <thinking> in LM Studio.

[-]

dreamai87@reddit

Thanks man bookmarked it. Very way to keep it

[-]

StrikeOner@reddit

Sorry, havent checked the model so far but i assume that the default qwen settings should work. You can find the settings at [https://llama-parampal.codecut.de/](https://llama-parampal.codecut.de/) . If they work better for you a feedback would be nice!