token counting

.gitattributes

@@ -63,3 +63,4 @@ redpajama-arxiv/ filter=lfs diff=lfs merge=lfs -text
 algebraic-stack/ filter=lfs diff=lfs merge=lfs -text
 arxiv/ filter=lfs diff=lfs merge=lfs -text
 open-web-math/ filter=lfs diff=lfs merge=lfs -text
+tokenizer.model filter=lfs diff=lfs merge=lfs -text

test_dataloader.py

@@ -2,18 +2,23 @@ from datasets import load_dataset
 from tqdm import tqdm
 import time
 
+import sentencepiece as spm
+
+s = spm.SentencePieceProcessor(model_file="tokenizer.model") # Llama tokenizer
+
 def main():
     for subset in ["arxiv", "open-web-math", "algebraic-stack"]:
         for split in ["train", "validation", "test"]:
             data = load_dataset("proof-pile-2.py", subset)[split]
             print(data)
 
+            num_toks = 0
             start = time.time()
             for x in tqdm(data):
-                pass
+                num_toks += len(s.encode(x['text']))
             total = time.time() - start
 
-            print(f"Traversed {subset}-{subset} in {total} seconds")
+            print(f"Traversed {num_toks:.5e} of {subset}-{split} in {total} seconds")
 
 if __name__=="__main__":
     main()

tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e556afd44213b6bd1be2b850ebbbd98f5481437a8021afaf58ee7fb1818d347
+size 499723