Datasets:

Modalities:
Tabular
Text
Formats:
parquet
Languages:
English
ArXiv:
DOI:
Libraries:
Datasets
Dask
License:
lvwerra HF staff commited on
Commit
c99ad79
1 Parent(s): 1d408f0

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +5 -1
README.md CHANGED
@@ -647,7 +647,11 @@ We compared 🍷 FineWeb with the following datasets:
647
 
648
  You will find these models on [this collection](https://huggingface.co/collections/HuggingFaceFW/ablation-models-662457b0d213e8c14fe47f32). We have uploaded checkpoints at every 1000 training steps. You will also find our full [evaluation results here](https://huggingface.co/datasets/HuggingFaceFW/fineweb/blob/main/eval_results.csv).
649
 
650
- [INSERT PLOT HERE]
 
 
 
 
651
 
652
  # Dataset card for 🍷 FineWeb
653
 
 
647
 
648
  You will find these models on [this collection](https://huggingface.co/collections/HuggingFaceFW/ablation-models-662457b0d213e8c14fe47f32). We have uploaded checkpoints at every 1000 training steps. You will also find our full [evaluation results here](https://huggingface.co/datasets/HuggingFaceFW/fineweb/blob/main/eval_results.csv).
649
 
650
+ <center>
651
+ <img src="https://huggingface.co/datasets/HuggingFaceFW/admin/resolve/main/fineweb-ablations.png" alt="ablations">
652
+ </center>
653
+
654
+ _Note:_ The plot is smoothed by averaging 5k steps in a rolling window.
655
 
656
  # Dataset card for 🍷 FineWeb
657