Revert "safer model pkl chunking (#37283)"

This reverts commit 5d54743d8b.

common/file_chunker.py

@@ -1,9 +1,8 @@
-import glob
 import math
 import os
 from pathlib import Path
 
-CHUNK_SIZE = 19 * 1024 * 1024  # 49MB, under GitHub's 50MB limit
+CHUNK_SIZE = 49 * 1024 * 1024  # 49MB, under GitHub's 50MB limit
 
 def get_chunk_name(name, idx, num_chunks):
   return f"{name}.chunk{idx+1:02d}of{num_chunks:02d}"
@@ -13,8 +12,6 @@ def get_chunk_paths(path, file_size):
   return [get_chunk_name(path, i, num_chunks) for i in range(num_chunks)]
 
 def chunk_file(path, num_chunks):
-  for old in glob.glob(f"{path}.chunk*"):
-    os.remove(old)
   with open(path, 'rb') as f:
     data = f.read()
   actual_num_chunks = max(1, math.ceil(len(data) / CHUNK_SIZE))
@@ -22,15 +19,13 @@ def chunk_file(path, num_chunks):
   for i in range(num_chunks):
     with open(get_chunk_name(path, i, num_chunks), 'wb') as f:
       f.write(data[i * CHUNK_SIZE:(i + 1) * CHUNK_SIZE])
-  os.remove(path)
 
 
 def read_file_chunked(path):
-  chunks = sorted(glob.glob(f"{path}.chunk*"))
-  if chunks:
-    expected = [get_chunk_name(path, i, len(chunks)) for i in range(len(chunks))]
-    assert chunks == expected, f"Chunk mismatch: {chunks} != {expected}"
-    return b''.join(Path(f).read_bytes() for f in chunks)
+  for num_chunks in range(1, 100):
+    if os.path.isfile(get_chunk_name(path, 0, num_chunks)):
+      files = [get_chunk_name(path, i, num_chunks) for i in range(num_chunks)]
+      return b''.join(Path(f).read_bytes() for f in files)
   if os.path.isfile(path):
     return Path(path).read_bytes()
   raise FileNotFoundError(path)