---
license: mit
---

These are the **first public interpreter models** trained
on a true reasoning model, and on **any model of this scale.** Because R1 is a
very large model and therefore difficult to run for most independent
researchers, we're also uploading SQL databases containing the max activating
examples for each feature.

## Model Information

This release contains two SAEs, one for general reasoning and one for math. After cloning our [demo repo](https://github.com/goodfire-ai/r1-interpretability/tree/main), you can load them
with the following snippet:

```python
from sae import load_math_sae
from huggingface_hub import hf_hub_download

file_path = hf_hub_download(
    repo_id=f"Goodfire/DeepSeek-R1-SAE-l37",
    filename=f"math/DeepSeek-R1-SAE-l37.pt",
    repo_type="model"
)
device = "cpu"
math_sae = load_math_sae(file_path, device)
```

The general reasoning SAE was trained on R1’s activations on our [custom
reasoning dataset](https://huggingface.co/datasets/Goodfire/r1-collect), and the second
used [OpenR1-Math](https://huggingface.co/datasets/open-r1/OpenR1-Math-220k), a
large dataset for mathematical reasoning. These datasets allow us to discover
the features that R1 uses to answer challenging problems that exercise its
reasoning chops.

Note: the original uploaded version of the logic SAE was incorrect; the correct version was uploaded on 4/17. If you have any difficulty running the SAEs, please reach out to us!