Merge pull request karpathy#10 from LaihoE/master

batch file write

Author: karpathy

data/openwebtext/prepare.py

@@ -54,12 +54,16 @@ def process(example):
     filename = os.path.join(os.path.dirname(__file__), f'{split}.bin')
     dtype = np.uint16 # (can do since enc.max_token_value == 50256 is < 2**16)
     arr = np.memmap(filename, dtype=dtype, mode='w+', shape=(arr_len,))
+    total_batches = 1024
 
-    print(f"writing {filename}...")
     idx = 0
-    for example in tqdm(dset):
-        arr[idx : idx + example['len']] = example['ids']
-        idx += example['len']
+    for batch_idx in tqdm(range(total_batches), desc=f'writing {filename}'):
+        # Batch together samples for faster write
+        batch = dset.shard(num_shards=total_batches, index=batch_idx, contiguous=True).with_format('numpy')
+        arr_batch = np.concatenate(batch['ids'])
+        # Write into mmap
+        arr[idx : idx + len(arr_batch)] = arr_batch
+        idx += len(arr_batch)
     arr.flush()
 
 # train.bin is ~17GB, val.bin ~8.5MB