QwQ

QwenLM/QwQ is ranked 6,834 among these 8 peers; mjhydri/BeatNet is ranked 6,832.

Unlike traditional instruction-tuned models, QwQ leverages advanced reasoning and critical thinking abilities to achieve superior performance on downstream tasks, especially those involving complex problem-solving. Our latest release, QwQ-32B, is a mid-sized model that competes effectively with top-tier reasoning models like DeepSeek-R1 and o1-mini, delivering robust and competitive results. **Note:** Please review the Usage Guidelines before deploying QwQ models, especially if you encounter **endless repetitions or significant performance issues**. Performance To reproduce the results

, please refer to our evaluation code. Quickstart with HuggingFace's transformers QwQ is based on Qwen2. 5, which has been in the latest Huggingface . We advise you to use the latest version of . With , you will encounter the following error: Here provides a code snippet with to show you how to load the tokenizer and model and how to generate contents. Usage Guidelines To achieve optimal performance, we recommend the following settings: 1. **Enforce Thoughtful Output**: Ensure the model starts with

"\ \n" to prevent generating empty thinking content, which can degrade output quality. If you use and set , this is already automatically implemented, but it may cause the response to lack the \ tag at the beginning. **Sampling Parameters**: **We recommend using Temperature=0. 95, MinP=0, TopK=40, and no repetition penalty for optimal performance. ** Do **NOT** use Greedy decoding under any circumstances! You can adjust the TopK value between 20 and 40 to balance filtering out rare token occurrences and

enhancing the diversity of the generated output. For supported frameworks, you can adjust the parameter between 0 and 2 to reduce endless repetitions. However, a higher value may occasionally result in language mixing and a slight decrease in performance. **No Thinking Content in History**: In multi-turn conversations, the historical model output should only include the final output part and does not need to include the thinking content. This feature is already implemented in . **Standardize Output Format**: We recommend using

prompts to standardize model outputs when benchmarking. **Math Problems**: Include "Please reason step by step, and put your final answer within \boxed{}. **Multiple-Choice Questions**: Add the following JSON structure to the prompt to standardize responses: "Please show your choice in the field with only the choice letter, e. **Handle Long Inputs**: For inputs exceeding 8,192 tokens, enable YaRN to improve the model's ability to capture long-sequence information effectively. cpp To run the Qwen/QwQ-32B-GGUF model

with Ollama, use the following command. cpp, you can run the model with the following command. This example uses the Q4_K_M quantization: You can also consult Unsloth's Guide to see if their approach meets your needs. ) Try QwQ with API If you face issues in deploying QwQ, we encourage you to test our API service provided by Alibaba Cloud Model Studio. Citation If you find our paper and code useful in your research, please consider giving a star

:star: and citation :pencil: :)