|
--- |
|
license: cc-by-3.0 |
|
datasets: |
|
- snats/datacomp_lists |
|
--- |
|
# Model Card: I want to break some laws too |
|
|
|
## Model Details |
|
|
|
- **Developer:** snats |
|
- **Model Date:** October 2024 |
|
- **Model Type:** Various checkpoints based on the DataComp small dataset |
|
- **Source:** https://snats.xyz/pages/articles/breaking_some_laws.html |
|
|
|
## Model Versions |
|
|
|
1. baseline |
|
2. breaking_0.1_trained |
|
3. breaking_0.2_trained |
|
4. breaking_0.3_trained |
|
5. breaking_0.5_trained |
|
6. breaking_0.7_trained |
|
7. breaking_0.9_trained |
|
8. minipile_style_only_txt |
|
9. minipile_style_trained |
|
10. top5_results |
|
|
|
## Model Description |
|
|
|
This repository contains multiple checkpoints for models trained as part of an experiment described in the blog post |
|
["I want to break some laws too"](https://snats.xyz/pages/articles/breaking_some_laws.html) on snats.xyz. |
|
The models are based on the DataComp small dataset and explore various data compositions. |
|
|
|
## Evaluation |
|
|
|
Each model version has its own evaluation results stored in its respective folder. |
|
|
|
## Additional Information |
|
|
|
- The models are stored as separate checkpoints under the same repository. |
|
- Each model folder contains its own evaluation results. |
|
- The `top5_results` folder contains evaluation results from the top 5 performing models, but not their weights. If you want, you can replicate it with |
|
the dataset found in [snats/datacomp_lists](https://huggingface.co/datasets/snats/datacomp_lists). |
|
- For detailed information on the training process, results, and insights, please refer to the original blog post. |
