Update README.md

README.md

@@ -110,3 +110,109 @@ configs:
   - split: test
     path: algebraic-stack/test-*
 ---
+
+![The header image of EleutherAI/proof-pile-2 edited to say "fixed"](./proofpile_fixed.png)
+
+The original [EleutherAI/proof-pile-2](https://huggingface.co/datasets/EleutherAI/proof-pile-2) dataset uses a custom python script and .jsonl.zst files, which some versions of the datasets library struggle with.
+
+This dataset contains the same data, subsets, and splits as [EleutherAI/proof-pile-2](https://huggingface.co/datasets/EleutherAI/proof-pile-2), converted into standard parquet format.
+
+Each subset and split was also shuffled so that directly training on the data is possible.
+
+Conversion was performed using the following script:
+```python
+
+import os
+import zstandard as zstd
+import json
+import pandas as pd
+from tqdm import tqdm
+
+import datasets
+import huggingface_hub as hf
+
+
+DATA_URL = "EleutherAI/proof-pile-2"
+SUBSETS = [
+    "algebraic-stack",
+    "arxiv",
+    "open-web-math"
+]
+SPLITS = [
+    "train",
+    "validation",
+    "test"
+]
+
+LOCAL_DIR = "./local_data/proof-pile-2"
+
+OUT_URL = 'aklein4/proof-pile-2-fixed'
+
+
+def download_data(
+    url: str,
+    subset: str,
+    split: str,
+):
+    hf.snapshot_download(
+        repo_id=url,
+        repo_type="dataset",
+        allow_patterns=[f"{subset}/{split}/*"],
+        local_dir=LOCAL_DIR,
+    )
+
+    return os.path.join(LOCAL_DIR, subset, split)
+
+
+def format_data(
+    url: str,
+    subset: str ,
+    split: str,
+):
+    
+    # download the data
+    folder = download_data(url, subset, split)
+
+    # get all files in the local dir
+    data_files = [
+        os.path.join(folder, f)
+        for f in os.listdir(folder)
+        if f.endswith(".zst")
+    ]
+
+    # read all of the .jsonl.zst files
+    examples = []
+    for file_path in tqdm(data_files):
+
+        with zstd.open(open(file_path, "rb"), "rt", encoding="utf-8") as f: 
+            for x in f.readlines():
+                examples.append(json.loads(x))
+
+    # get the dataset
+    df = pd.DataFrame(examples)
+    dataset = datasets.Dataset.from_pandas(df)
+    dataset = dataset.shuffle(seed=42)
+
+    dataset.push_to_hub(
+        OUT_URL,
+        config_name=subset,
+        split=split,
+        private=False
+    )
+
+
+def main():
+
+    for subset in SUBSETS:
+        for split in SPLITS:
+            
+            format_data(
+                DATA_URL,
+                subset,
+                split,
+            )
+
+
+if __name__ == "__main__":
+    main()
+```