Add docstrings to some modules and classes (#100)

cacheflow/model_executor/weight_utils.py

@@ -1,3 +1,4 @@
+"""Utilities for downloading and initializing model weights."""
 import filelock
 import glob
 import json
@@ -106,5 +107,12 @@ def initialize_dummy_weights(
     low: float = -1e-3,
     high: float = 1e-3,
 ) -> None:
+    """Initialize model weights with random values.
+
+    The model weights must be randomly initialized for accurate performance
+    measurements. Additionally, the model weights should not cause NaNs in the
+    forward pass. We empirically found that initializing the weights with
+    values between -1e-3 and 1e-3 works well for most models.
+    """
     for param in model.state_dict().values():
         param.data.uniform_(low, high)