diff --git a/hyak_scripts/filter.py b/hyak_scripts/filter.py
new file mode 100644
index 0000000..1c0516e
--- /dev/null
+++ b/hyak_scripts/filter.py
@@ -0,0 +1,25 @@
+import dask
+import dask.dataframe as dd
+import numpy as np 
+import pyarrow.parquet as pq
+import pandas as pd
+
+metadata_path = '/mmfs1/gscratch/krishna/mayank/dfn/dfn-medium/metadata'
+
+indices = np.load('/mmfs1/gscratch/krishna/mayank/dfn/indicies/datacomp_medium_dfn_20m_inds.npy')
+all_uids = set([f'{uid[0]:016x}{uid[1]:016x}' for uid in indices])
+
+fp = pq.read_table(
+    source=metadata_path,
+    use_threads=True,
+    filters=[('uid', 'in', all_uids)]
+)
+schema = pq.read_schema("/mmfs1/gscratch/krishna/mayank/dfn/dfn-medium/metadata/0a4a1e10352ec4366858927c33873cdc.parquet")
+df = fp.to_pandas()
+
+# save df to disk
+dd.from_pandas(df, chunksize=500000).to_parquet(
+    '/mmfs1/gscratch/krishna/mayank/dfn/dfn-medium/metadata_filtered', 
+    schema=schema
+    )
+
diff --git a/hyak_scripts/upload.py b/hyak_scripts/upload.py
new file mode 100644
index 0000000..b602bae
--- /dev/null
+++ b/hyak_scripts/upload.py
@@ -0,0 +1,40 @@
+import img2dataset
+import argparse
+from cloudpathlib import CloudPath
+
+def upload(source, destination):
+    img2dataset.download(
+        processes_count=16,
+        thread_count=128,
+        image_size=512,
+        resize_mode="keep_ratio_largest",
+        url_list = str(source),
+        resize_only_if_bigger=True,
+        encode_format="jpg",
+        output_format="webdataset",
+        retries=3,
+        enable_wandb=False,
+        wandb_project="dataupload",
+        skip_reencode=True,
+        output_folder=destination,
+        input_format="parquet",
+        url_col="url",
+        caption_col="text",
+        number_sample_per_shard=10000,
+        distributor="multiprocessing", 
+        save_additional_columns=["uid"],
+        oom_shard_count=8,
+        bbox_col="face_bboxes",
+    )
+    print("download complete!!")
+    
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--source", type=str, required=True)
+    parser.add_argument("--destination", type=str, required=True)
+    args = parser.parse_args()
+    # destination = CloudPath(args.destination)
+    upload(args.source, args.destination)
+
+# python3 upload.py --source /mmfs1/gscratch/krishna/mayank/dfn/dfn-medium/metadata_filtered --destination gs://dfn_medium    
\ No newline at end of file