huggingface/open-r1

View on GitHub

Fully open reproduction of DeepSeek-R1

Stars: 26.5k
Forks: 2.5k
Open beginner issues: 0
Indexed issues: 291

Dominant language: Python
License: Apache-2.0
Last GitHub push: Apr 2, 2026
Latest indexed: Sep 17, 2026
Contributing guide: No contributing guide
Code of conduct: No code of conduct
Beginner labels: No beginner labels indexed
PR merge metrics: No merged PRs in 30d

Open All

291 open issues indexed

Evaluate GRPO vs. other RL algorithms Open

huggingface/open-r1#11 · 8 comments · 5 reactions · 0 assignees · Indexed Sep 17, 2026
Wondering would consider support muli-modal Open

huggingface/open-r1#12 · 5 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
How to contribute Open

huggingface/open-r1#23 · 12 comments · 162 reactions · 0 assignees · Indexed Sep 17, 2026
Datasets for math Open

huggingface/open-r1#24 · 19 comments · 3 reactions · 0 assignees · Indexed Sep 17, 2026
Datasets for code Open

huggingface/open-r1#28 · 29 comments · 7 reactions · 0 assignees · Indexed Sep 17, 2026
Datasets for Medicine Open

huggingface/open-r1#31 · 3 comments · 14 reactions · 0 assignees · Indexed Sep 17, 2026
Function calling at thinking time Open

huggingface/open-r1#34 · 8 comments · 1 reaction · 0 assignees · Indexed Sep 17, 2026
what about open-v3? Open

huggingface/open-r1#35 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Wondering why no format reward? Open

huggingface/open-r1#45 · 3 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
how to train on MultiNode MultiGPU Open

huggingface/open-r1#46 · 2 comments · 9 reactions · 0 assignees · Indexed Sep 17, 2026
Crazy VRAM usage with longer prompts Open

huggingface/open-r1#47 · 16 comments · 3 reactions · 0 assignees · Indexed Sep 17, 2026
[Willing to Contribute] Integrate SGLang into open-r1 Open

huggingface/open-r1#48 · 0 comments · 12 reactions · 0 assignees · Indexed Sep 17, 2026
Datasets for law Open

huggingface/open-r1#50 · 12 comments · 2 reactions · 0 assignees · Indexed Sep 17, 2026
Plan of attack uses R1 Zero for supervised fine-tuning in step 3. Open

huggingface/open-r1#53 · 1 comment · 6 reactions · 0 assignees · Indexed Sep 17, 2026
How to supervise non-math data? Open

huggingface/open-r1#56 · 4 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Datasets for linguistics Open

huggingface/open-r1#60 · 1 comment · 1 reaction · 0 assignees · Indexed Sep 17, 2026
Datasets for fine arts Open

huggingface/open-r1#61 · 5 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Could you provide the model trained by GRPO? Open

huggingface/open-r1#64 · 0 comments · 2 reactions · 0 assignees · Indexed Sep 17, 2026
We Need a Better Training Pipeline: GRPO Trainer Struggles with Long Completion Lengths on H100x8 Open

huggingface/open-r1#65 · 4 comments · 12 reactions · 0 assignees · Indexed Sep 17, 2026
FP8 training Open

huggingface/open-r1#68 · 1 comment · 1 reaction · 0 assignees · Indexed Sep 17, 2026
ValueError: please provide at least one prompt Open

huggingface/open-r1#69 · 5 comments · 3 reactions · 0 assignees · Indexed Sep 17, 2026
Why the average power usage of GRPO training is at 50%? Open

huggingface/open-r1#83 · 0 comments · 1 reaction · 0 assignees · Indexed Sep 17, 2026
tokenizer.chat_template is not set and no template argument was passed Open

huggingface/open-r1#85 · 4 comments · 5 reactions · 0 assignees · Indexed Sep 17, 2026
Issue: Potential Inflation of Reward Metrics for Unparseable Gold Solutions Open

huggingface/open-r1#86 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
As the inference on local device is unaffordable for most people. Open

huggingface/open-r1#88 · 4 comments · 1 reaction · 0 assignees · Indexed Sep 17, 2026
evaluate code for livecodebench, GPQA and Codeforces Open

huggingface/open-r1#89 · 3 comments · 3 reactions · 0 assignees · Indexed Sep 17, 2026
Bugs in evaluation? Cannot reproduce the numbers for deepseek models Open

huggingface/open-r1#97 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Any open tasks for people who don't have access to compute? Open

huggingface/open-r1#98 · 0 comments · 10 reactions · 0 assignees · Indexed Sep 17, 2026
What is the compute needed for GRPO for 7B R1-Distill model? Open

huggingface/open-r1#100 · 14 comments · 4 reactions · 0 assignees · Indexed Sep 17, 2026
Init file imports Open

huggingface/open-r1#101 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
installation error: No matching distribution found for bitsandbytes>=0.43.0 Open

huggingface/open-r1#108 · 4 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
What is the GPU resource required to run Open-R1 (Deepseek-R1) locally? Open

huggingface/open-r1#113 · 2 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
crazy idea, train and inference via cpu, is it available? Open

huggingface/open-r1#114 · 3 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Lighteval is incompatible with torch v2.5.1 Open

huggingface/open-r1#122 · 1 comment · 0 reactions · 0 assignees · Indexed Sep 17, 2026
An dataset error? Open

huggingface/open-r1#126 · 3 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:1 and cuda:0! Open

huggingface/open-r1#129 · 6 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
About reward signal design in GRPO Open

huggingface/open-r1#135 · 5 comments · 3 reactions · 0 assignees · Indexed Sep 17, 2026
AttributeError: 'Qwen2ForCausalLM' object has no attribute 'optimizer' during DRPO training Open

huggingface/open-r1#145 · 9 comments · 12 reactions · 0 assignees · Indexed Sep 17, 2026
[GRPO Training Errors] 1.(Qwen2ForCausalLM object has no attribute 'optimizer' ); 2.(ZeroDivisionError) Open

huggingface/open-r1#146 · 24 comments · 5 reactions · 0 assignees · Indexed Sep 17, 2026
TypeError: SFTTrainer.create_model_card() got an unexpected keyword argument 'finetuned_from' Open

huggingface/open-r1#148 · 5 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
GRPO issue Open

huggingface/open-r1#152 · 3 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Enable Apple Silicon Support with PyTorch’s MPS Backend for Training on M*-{Max,Ultra} Chips Open

huggingface/open-r1#153 · 1 comment · 3 reactions · 0 assignees · Indexed Sep 17, 2026
Is the pipeline described in Step 3 diagram in the README.md correct? Open

huggingface/open-r1#154 · 0 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Continual Training for DeepSeek R1 Open

huggingface/open-r1#156 · 0 comments · 1 reaction · 0 assignees · Indexed Sep 17, 2026
Slow Training Time in GRPO Open

huggingface/open-r1#161 · 2 comments · 7 reactions · 0 assignees · Indexed Sep 17, 2026
RuntimeError: CUDA driver error: invalid argument Open

huggingface/open-r1#162 · 5 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Error in make evaluate: TypeError: VLLMModelConfig.__init__() got an unexpected keyword argument '' Open

huggingface/open-r1#163 · 1 comment · 1 reaction · 0 assignees · Indexed Sep 17, 2026
Is Data Preprocessing Recommended for AI-MO/NuminaMath-TIR? Open

huggingface/open-r1#165 · 2 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
Thinking Duration Controller Open

huggingface/open-r1#167 · 2 comments · 0 reactions · 0 assignees · Indexed Sep 17, 2026
GRPO training error related to sympy/math expressions Open

huggingface/open-r1#168 · 4 comments · 2 reactions · 0 assignees · Indexed Sep 17, 2026

huggingface/open-r1

Get new issues in your inbox