Reduce the complete storage & reading thresholds for int and float from 1024 to 64 (#180)

* CONTRIBUTE_TUTORIAL_cn.md

* Handle big int tensors by converting to sparse COO

* Update utils

* Update utils

* Update utils

* Update utils

Author: JewelRoam

graph_net/torch/utils.py

@@ -52,7 +52,7 @@ def process_tensor(tensor):
 
         info = tensor_info(tensor)
         if tensor.dtype in [torch.int8, torch.int16, torch.int32, torch.int64]:
-            if tensor.numel() < 1024:
+            if tensor.numel() < 64:
                 return {
                     "type": "small_int_tensor",
                     "data": tensor.clone(),
@@ -65,7 +65,7 @@ def process_tensor(tensor):
                     "max_val": tensor.max().item(),
                     "info": info,
                 }
-        elif tensor.numel() < 1024:
+        elif tensor.numel() < 64:
             return {"type": "small_tensor", "data": tensor.clone(), "info": info}
         else:
             return {"type": "random_tensor", "info": info}
@@ -80,7 +80,7 @@ def process_tensor(tensor):
     def handle_named_tensors(tensor):
         info = tensor_info(tensor)
         if tensor.dtype in [torch.int8, torch.int16, torch.int32, torch.int64]:
-            if tensor.numel() < 1024:
+            if tensor.numel() < 64:
                 return {
                     "info": info,
                     "data": tensor.clone(),
@@ -93,7 +93,7 @@ def handle_named_tensors(tensor):
                     "max_val": tensor.max().item(),
                     "type": "big_int_tensor_by_range",
                 }
-        if tensor.numel() < 1024:
+        if tensor.numel() < 64:
             return {"info": info, "data": tensor.clone(), "type": "small_tensor"}
         else:
             return {"info": info, "data": None, "type": "random_tensor"}