open-thoughts/OpenThinker-7B

1---2library_name: transformers3license: apache-2.04base_model: Qwen/Qwen2.5-7B-Instruct5tags:6- llama-factory7- full8- generated_from_trainer9model-index:10- name: OpenThinker-7B11  results: []12datasets:13- open-thoughts/open-thoughts-114k14---15 16<p align="center">17    <img src="https://huggingface.co/datasets/open-thoughts/open-thoughts-114k/resolve/main/open_thoughts.png" width="50%">18</p>19 20> [!NOTE]21> We have released a paper for OpenThoughts! See our paper [here](https://arxiv.org/abs/2506.04178).22 23# OpenThinker-7B24 25This model is a fine-tuned version of [Qwen/Qwen2.5-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-7B-Instruct) on the 26[OpenThoughts-114k dataset](https://huggingface.co/datasets/open-thoughts/OpenThoughts-114k) dataset.27 28The dataset is derived by distilling DeepSeek-R1 using the [data pipeline available on github](https://github.com/open-thoughts/open-thoughts). 29More info about the dataset can be found on the dataset card at [OpenThoughts-114k dataset](https://huggingface.co/datasets/open-thoughts/open-thoughts-114k).30 31This model improves upon the [Bespoke-Stratos-7B model](https://huggingface.co/bespokelabs/Bespoke-Stratos-7B), which used 17k examples ([Bespoke-Stratos-17k dataset](https://huggingface.co/datasets/bespokelabs/Bespoke-Stratos-17k)).32The numbers reported in the table below are evaluated with our open-source tool [Evalchemy](https://github.com/mlfoundations/Evalchemy).33 34|                             | AIME24   | MATH500 | GPQA-Diamond | LCBv2 Easy  | LCBv2 Medium  | LCBv2 Hard  | LCBv2 All  |35| --------------------------- | -------- | ------- | ------------ | ----------- | ------------- | ----------- | ---------- |36| OpenThinker-7B              | 31.3     | 83.0    | 42.4         | 75.3        | 28.6          | 6.5         | 39.9       |37| Bespoke-Stratos-7B          | 22.7     | 79.6    | 38.9         | 71.4        | 25.2          | 0.8         | 35.8       |38| DeepSeek-R1-Distill-Qwen-7B | 60       | 88.2    | 46.9         | 79.7        | 45.1          | 14.6        | 50.1       |39| gpt-4o-0513                 | 8.7       | 75.8    | 46.5         | 87.4        | 42.7          | 8.9         | 50.5       |40| o1-mini                     | 64       | 85.6    | 60           | 92.8        | 74.7          | 39.8        | 72.8       |41 42We are fully open-source. Our [model weights](https://huggingface.co/open-thoughts), [datasets](https://huggingface.co/open-thoughts), [data generation code](https://github.com/open-thoughts/open-thoughts), [evaluation code](https://github.com/mlfoundations/Evalchemy), and [training code](https://github.com/hiyouga/LLaMA-Factory) are all publicly available. 43 44|  | Open Weights | Open Data | Open Code | 45|--|--------------|-----------| --------- |46|OpenThinker-7B|✅|[✅](https://huggingface.co/datasets/open-thoughts/OpenThoughts-114k)|[✅](https://github.com/open-thoughts/open-thoughts) |47|Bespoke-Stratos-7B|✅|[✅](https://huggingface.co/datasets/bespokelabs/Bespoke-Stratos-17k)|[✅](https://github.com/bespokelabsai/curator/tree/main/examples/bespoke-stratos-data-generation)|48|DeepSeek-R1-Distill-Qwen-7B|✅|❌|❌|49|gpt-4o-0513|❌|❌|❌|❌|50|o1-mini|❌|❌|❌|❌|51 52 53## Intended uses & limitations54 55Apache 2.0 License56 57 58## Training procedure59 60We used four 8xH100 nodes to train the model for 20 hours.61 62### Training hyperparameters63 64The following hyperparameters were used during training:65- learning_rate: 1e-0566- train_batch_size: 167- eval_batch_size: 868- seed: 4269- distributed_type: multi-GPU70- num_devices: 3271- gradient_accumulation_steps: 372- total_train_batch_size: 9673- total_eval_batch_size: 25674- optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments75- lr_scheduler_type: cosine76- lr_scheduler_warmup_ratio: 0.177- num_epochs: 3.078 79### Framework versions80 81- Transformers 4.46.182- Pytorch 2.3.083- Datasets 3.1.084- Tokenizers 0.20.385 86More info can be found in our repository: [https://github.com/open-thoughts/open-thoughts](https://github.com/open-thoughts/open-thoughts).87 88# Links89- 📝 [OpenThoughts Paper](https://arxiv.org/abs/2506.04178)90- 📊 [Open Thoughts Launch Blog Post](https://www.open-thoughts.ai/blog/launch)91- 💻 [Open Thoughts GitHub Repository](https://github.com/open-thoughts/open-thoughts)92- 🧠 [OpenThoughts-114k dataset](https://huggingface.co/datasets/open-thoughts/OpenThoughts-114k)93- 🤖 [OpenThinker-7B model](https://huggingface.co/open-thoughts/OpenThinker-7B) - this model.94- 📊 [Bespoke-Stratos Blog Post](https://www.bespokelabs.ai/blog/bespoke-stratos-the-unreasonable-effectiveness-of-reasoning-distillation)95- 🧠 [Bespoke-Stratos-17k dataset](https://huggingface.co/datasets/bespokelabs/Bespoke-Stratos-17k)96- 🤖 [Bespoke-Stratos-32B model](https://huggingface.co/bespokelabs/Bespoke-Stratos-32B)97- 🤖 [Bespoke-Stratos-7B model](https://huggingface.co/bespokelabs/Bespoke-Stratos-7B)98 99# Citation100```101@misc{guha2025openthoughtsdatarecipesreasoning,102  title={OpenThoughts: Data Recipes for Reasoning Models}, 103  author={Etash Guha and Ryan Marten and Sedrick Keh and Negin Raoof and Georgios Smyrnis and Hritik Bansal and Marianna Nezhurina and Jean Mercat and Trung Vu and Zayne Sprague and Ashima Suvarna and Benjamin Feuer and Liangyu Chen and Zaid Khan and Eric Frankel and Sachin Grover and Caroline Choi and Niklas Muennighoff and Shiye Su and Wanjia Zhao and John Yang and Shreyas Pimpalgaonkar and Kartik Sharma and Charlie Cheng-Jie Ji and Yichuan Deng and Sarah Pratt and Vivek Ramanujan and Jon Saad-Falcon and Jeffrey Li and Achal Dave and Alon Albalak and Kushal Arora and Blake Wulfe and Chinmay Hegde and Greg Durrett and Sewoong Oh and Mohit Bansal and Saadia Gabriel and Aditya Grover and Kai-Wei Chang and Vaishaal Shankar and Aaron Gokaslan and Mike A. Merrill and Tatsunori Hashimoto and Yejin Choi and Jenia Jitsev and Reinhard Heckel and Maheswaran Sathiamoorthy and Alexandros G. Dimakis and Ludwig Schmidt},104  year={2025},105  eprint={2506.04178},106  archivePrefix={arXiv},107  primaryClass={cs.LG},108  url={https://arxiv.org/abs/2506.04178}, 109}110```111