Key Technical Specifications

Best Practices

• Use `temperature=0.6` and `top_p=0.95` to avoid endless repetitions and hallucinations.
• Utilize the full context window - with 128K tokens available, provide comprehensive problem descriptions and relevant background information.
• Set `reasoning_format` to `parsed` with to handle the missing first `<think>` token in QwQ-32B output.
• For multi-turn conversations, include only the final output from previous turns in history, not the thinking content.
• Prompt the model to be concise when needed - the model tends to produce extensive reasoning chains.
• Increase `max_completion_tokens` to give the model sufficient space to complete its reasoning without truncation.
• If reasoning chains are critical, prompt the model to avoid using Chinese characters in its output (this is normal behavior).
• Take advantage of QwQ-32B's strong tool use and function calling capabilities for agentic applications.
• If the model provides thinking without reaching a final answer, try prompting for conciseness or rerun your query.

Get Started with Qwen/QwQ-32B

Experience the world's fastest breakthrough reasoning capabilities with qwen-qwq-32b on Groq:

pip install groq

from groq import Groq
client = Groq()
completion = client.chat.completions.create(
    model="qwen-qwq-32b",
    messages=[
        {
            "role": "user",
            "content": "Explain why fast inference is critical for reasoning models"
        }
    ]
)
print(completion.choices[0].message.content)