Update prediction data with ngram processor

README.md

@@ -5,14 +5,16 @@ configs:
   data_files:
   - split: train
     path: bytelevel/*.parquet
-- config_name: bytelevel-subset
-  data_files:
-  - split: train
-    path: bytelevel-subset/*.parquet
 - config_name: bytelevel-llm-data
   data_files:
   - split: ngram
     path: ngram/train-*
+  - split: train
+    path: bytelevel-llm-data/ngram/train-*
+- config_name: bytelevel-subset
+  data_files:
+  - split: train
+    path: bytelevel-subset/*.parquet
 language:
 - en
 tags:
@@ -20,6 +22,29 @@ tags:
 pretty_name: FineWebEDU 20B
 size_categories:
 - 10B<n<100B
+dataset_info:
+  config_name: bytelevel-llm-data
+  features:
+  - name: id
+    dtype: string
+  - name: input_ids
+    sequence: int32
+  - name: num_tokens
+    dtype: int64
+  - name: Entropy
+    sequence: float64
+  - name: Surprisal
+    sequence: float64
+  - name: Space Probability
+    sequence: float64
+  - name: EOS Probability
+    sequence: float64
+  splits:
+  - name: train
+    num_bytes: 226886
+    num_examples: 2
+  download_size: 160553
+  dataset_size: 226886
 ---
 
 # FineWebEDU 20B

bytelevel-llm-data/ngram/train-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a88c283af56573dc5992c8510e5d2825796b493208728a410a8e464e297aecd5
+size 160553