Ultra-fast dataset upload with pre-encoded bytes

README.md

@@ -99,17 +99,17 @@ dataset_info:
   - name: target_hash
     dtype: string
   splits:
-  - name: train
-    num_bytes: 575397253
-    num_examples: 12736
   - name: val
     num_bytes: 48223802
     num_examples: 1040
   - name: total
     num_bytes: 1179312913
     num_examples: 26054
-  download_size: 10822066824
-  dataset_size: 1809520561
+  - name: train
+    num_bytes: 750652393
+    num_examples: 16685
+  download_size: 11189939046
+  dataset_size: 1984775701
 configs:
 - config_name: comparison
   data_files:

streaming/train-00000-of-00002.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:dc7097995951df8d856ad3a82038e7dca3d64a57865e00e7d9f80ee7ac841204
-size 205742504
+oid sha256:6d3501f0de33ffc09fa5f8b3b6f5865446f2546d38ddc3ef3b5f6a89408bc24f
+size 180173669

streaming/train-00001-of-00002.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a53cdb04d2d0740d8d51fde73d133a97367ee99e351af473b97589cde526e36a
-size 216028334
+oid sha256:1752b8846c9a253a033453a38e8fc6b13ec69ba94e77a92869372911b23895b3
+size 187698553