runeblaze

↗ HN profile [ 80.8 ms ] full profile
Karma
0
Created
()
Submissions
8,172,457
  1. In large-scale ML training, shuffling datasets with over 1 billion data points (e.g., LAION-5B) is common. Typical methods like `np.random.shuffle` on an indices array, `df.sample(frac=1)` on the entire dataset, or…