Add more comments in sharded_tbes_weights_spec (pytorch#1596)

Summary:
Pull Request resolved: pytorch#1596

As titled.

Reviewed By: henrylhtsang

Differential Revision: D52341591

fbshipit-source-id: 168c6345f2b236fe774d4c0c8d3a5a5a7bff17c4

torchrec/distributed/quant_state.py

@@ -426,6 +426,8 @@ def sharded_tbes_weights_spec(
     # 	"ebc.tbes.1.1.table_1.weight_qscale":WeightSpec("ebc.embedding_bags.table_1.weight_qscale", [500, 0], [500, 2])
     # 	"ebc.tbes.1.1.table_1.weight_qbias":WeightSpec("ebc.embedding_bags.table_1.weight_qbias", [500, 0], [500, 2])
     # }
+    # In the format of ebc.tbes.i.j.table_k.weight, where i is the index of the TBE, j is the index of the embedding bag within TBE i, k is the index of the original table set in the ebc embedding_configs
+    # e.g. ebc.tbes.1.1.table_1.weight, it represents second embedding bag within the second TBE. This part of weight is from a shard of table_1
 
     ret: Dict[str, WeightSpec] = {}
     for module_fqn, module in sharded_model.named_modules():