--- license: apache-2.0 library_name: transformers base_model: qihoo360/TinyR1-32B-Preview tags: - llama-cpp - gguf-my-repo --- # cassettesgoboom/TinyR1-32B-Preview-Q3_K_L-GGUF ![image/png](https://cdn-uploads.huggingface.co/production/uploads/671671451e5b1ae705452628/SuKWNMwaQTR6cfpNOR4Sw.png) This model was converted to GGUF format from [`qihoo360/TinyR1-32B-Preview`](https://huggingface.co/qihoo360/TinyR1-32B-Preview) using llama.cpp via the ggml.ai's [GGUF-my-repo](https://huggingface.co/spaces/ggml-org/gguf-my-repo) space. Refer to the [original model card](https://huggingface.co/qihoo360/TinyR1-32B-Preview) for more details on the model. Original Model card: --- license: apache-2.0 library_name: transformers base_model: - deepseek-ai/DeepSeek-R1-Distill-Qwen-32B --- **Model Name**: Tiny-R1-32B-Preview **Title**: SuperDistillation Achieves Near-R1 Performance with Just 5% of Parameters. # Introduction We introduce our first-generation reasoning model, Tiny-R1-32B-Preview, which outperforms the 70B model Deepseek-R1-Distill-Llama-70B and nearly matches the full R1 model in math. ## Evaluation | Model | Math (AIME 2024) | Coding (LiveCodeBench) | Science (GPQA-Diamond) | | ------------------------------- | ------------------- | ----------------------- | ---------------------- | | Deepseek-R1-Distill-Qwen-32B | 72.6 | 57.2 | 62.1 | | Deepseek-R1-Distill-Llama-70B | 70.0 | 57.5 | 65.2 | | Deepseek-R1 | 79.8 | 65.9 | 71.5 | | Tiny-R1-32B-Preview (Ours) | 78.1 | 61.6 | 65.0 All scores are reported as pass@1. For AIME 2024, we sample 16 responses, and for GPQA-Diamond, we sample 4 responses, both using average overall accuracy for stable evaluation. ## Approach | Model | Math (AIME 2024) | Coding (LiveCodeBench) | Science (GPQA-Diamond) | | ------------------------------- | ------------------- | ----------------------- | ---------------------- | | Math-Model (Ours) | 73.1 | - | - | | Code-Model (Ours) | - | 63.4 | - | | Science-Model (Ours) | - | - | 64.5 | | Tiny-R1-32B-Preview (Ours) | 78.1 | 61.6 | 65.0 We applied supervised fine-tuning (SFT) to Deepseek-R1-Distill-Qwen-32B across three target domains—Mathematics, Code, and Science — using the [360-LLaMA-Factory](https://github.com/Qihoo360/360-LLaMA-Factory/) training framework to produce three domain-specific models. We used questions from open-source data as seeds, and used DeepSeek-R1 to generate responses for mathematics, coding, and science tasks separately, creating specialized models for each domain. Building on this, we leveraged the Mergekit tool from the Arcee team to combine multiple models, creating Tiny-R1-32B-Preview, which demonstrates strong overall performance. ## Data #### 1. Math 58.3k CoT trajectories from [open-r1/OpenR1-Math-220k](https://huggingface.co/datasets/open-r1/OpenR1-Math-220k), default subset #### 2. Coding 19k CoT trajectories [open-thoughts/OpenThoughts-114k](https://huggingface.co/datasets/open-thoughts/OpenThoughts-114k), coding subset #### 3. Science We used R1 to generate 8 CoT trajectories on 7.6k seed examples, and got 60.8k CoT trajectories in total; the seed examples are as follows: - 2.7k seed examples from [simplescaling/data_ablation_full59K](https://huggingface.co/datasets/simplescaling/data_ablation_full59K), science and health science subset - 4.9k seed examples from [open-thoughts/OpenThoughts-114k](https://huggingface.co/datasets/open-thoughts/OpenThoughts-114k), science subset ## Open Source Plan We will publish a technical report as soon as possible and open-source our training and evaluation code, selected training data, and evaluation logs. Having benefited immensely from the open-source community, we are committed to giving back in every way we can. ## Contributors *360 Team:* Lin Sun, Guangxiang Zhao, Xiaoqi Jian, Weihong Lin, Yongfu Zhu, Change Jia, Linglin Zhang, Jinzhu Wu, Sai-er Hu, Xiangzheng Zhang *PKU Team:* Yuhan Wu, Zihan Jiang, Wenrui Liu, Junting Zhou, Bin Cui, Tong Yang ## Citation ``` @misc{tinyr1proj, title={SuperDistillation Achieves Near-R1 Performance with Just 5% of Parameters.}, author={TinyR1 Team}, year={2025}, eprint={}, archivePrefix={}, primaryClass={}, url={https://huggingface.co/qihoo360/TinyR1-32B-Preview}, } ``` ## Use with llama.cpp Install llama.cpp through brew (works on Mac and Linux) ```bash brew install llama.cpp ``` Invoke the llama.cpp server or the CLI. ### CLI: ```bash llama-cli --hf-repo cassettesgoboom/TinyR1-32B-Preview-Q3_K_L-GGUF --hf-file tinyr1-32b-preview-q3_k_l.gguf -p "The meaning to life and the universe is" ``` ### Server: ```bash llama-server --hf-repo cassettesgoboom/TinyR1-32B-Preview-Q3_K_L-GGUF --hf-file tinyr1-32b-preview-q3_k_l.gguf -c 2048 ``` Note: You can also use this checkpoint directly through the [usage steps](https://github.com/ggerganov/llama.cpp?tab=readme-ov-file#usage) listed in the Llama.cpp repo as well. Step 1: Clone llama.cpp from GitHub. ``` git clone https://github.com/ggerganov/llama.cpp ``` Step 2: Move into the llama.cpp folder and build it with `LLAMA_CURL=1` flag along with other hardware-specific flags (for ex: LLAMA_CUDA=1 for Nvidia GPUs on Linux). ``` cd llama.cpp && LLAMA_CURL=1 make ``` Step 3: Run inference through the main binary. ``` ./llama-cli --hf-repo cassettesgoboom/TinyR1-32B-Preview-Q3_K_L-GGUF --hf-file tinyr1-32b-preview-q3_k_l.gguf -p "The meaning to life and the universe is" ``` or ``` ./llama-server --hf-repo cassettesgoboom/TinyR1-32B-Preview-Q3_K_L-GGUF --hf-file tinyr1-32b-preview-q3_k_l.gguf -c 2048 ```