fixed error when too many model weights are sent to frontend

src/models_builder/gnn_constructor.py

@@ -108,7 +108,7 @@ def get_hash(self):
         gnn_name_hash = hash_data_sha256(json_object.encode('utf-8'))
         return gnn_name_hash
 
-    def get_full_info(self):
+    def get_full_info(self, tensor_size_limit=None):
         """ Get available info about model for frontend
         """
         # FIXMe architecture and weights can be not accessible
@@ -118,7 +118,7 @@ def get_full_info(self):
         except (AttributeError, NotImplementedError):
             pass
         try:
-            result["weights"] = self.get_weights()
+            result["weights"] = self.get_weights(tensor_size_limit=tensor_size_limit)
         except (AttributeError, NotImplementedError):
             pass
         try:
@@ -165,9 +165,8 @@ def get_neurons(self):
                 neurons.append(n_neurons)
         return neurons
 
-    def get_weights(self):
-        """
-        Get model weights calling torch.nn.Module.state_dict() to draw them on the frontend.
+    def get_weights(self, tensor_size_limit=None):
+        """ Get model weights calling torch.nn.Module.state_dict() to draw them on the frontend.
         """
         try:
             state_dict = self.state_dict()
@@ -185,8 +184,15 @@ def get_weights(self):
 
             k = sub_keys[-1]
             if type(value) == UninitializedParameter:
-                continue
-            part[k] = value.numpy().tolist()
+                part[k] = '?'
+            else:
+                size = 1
+                for dim in value.shape:
+                    size *= dim
+                if tensor_size_limit and size > tensor_size_limit:  # Tensor is too big - return just its shape
+                    part[k] = 'x'.join(str(d) for d in value.shape)
+                else:
+                    part[k] = value.numpy().tolist()
         return model_data
 
 

web_interface/back_front/model_blocks.py

@@ -17,6 +17,8 @@
 from web_interface.back_front.block import Block, WrapperBlock
 from web_interface.back_front.utils import WebInterfaceError, json_dumps, get_config_keys
 
+TENSOR_SIZE_LIMIT = 1024  # Max size of weights tensor we sent to frontend
+
 
 class ModelWBlock(WrapperBlock):
     def __init__(self, name, blocks, *args, **kwargs):
@@ -58,7 +60,7 @@ def _submit(self):
 
         self._object = self.model_manager
         self._result = self._object.get_full_info()
-        self._result.update(self._object.gnn.get_full_info())
+        self._result.update(self._object.gnn.get_full_info(tensor_size_limit=TENSOR_SIZE_LIMIT))
 
     def get_index(self):
         """ Get all available models with respect to current dataset
@@ -71,9 +73,8 @@ def get_index(self):
         keys_list, full_keys_list, dir_structure, _ = DataInfo.take_keys_etc_by_prefix(
             prefix=("data_root", "data_prepared")
         )
-        values_info = DataInfo.values_list_by_path_and_keys(path=path,
-                                                            full_keys_list=full_keys_list,
-                                                            dir_structure=dir_structure)
+        values_info = DataInfo.values_list_by_path_and_keys(
+            path=path, full_keys_list=full_keys_list, dir_structure=dir_structure)
         ps = index.filter(dict(zip(keys_list, values_info)))
         return [ps.to_json(), json_dumps(info)]
 
@@ -105,7 +106,7 @@ def _finalize(self):
 
     def _submit(self):
         self._object = FrameworkGNNConstructor(self.model_config)
-        self._result = self._object.get_full_info()
+        self._result = self._object.get_full_info(tensor_size_limit=TENSOR_SIZE_LIMIT)
 
 
 class ModelCustomBlock(Block):
@@ -140,7 +141,7 @@ def _submit(self):
         assert cm_path
 
         self._object = UserCodeInfo.take_user_model_obj(cm_path, self.model_name["model"])
-        self._result = self._object.get_full_info()
+        self._result = self._object.get_full_info(tensor_size_limit=TENSOR_SIZE_LIMIT)
 
     def get_index(self):
         """ Get all available models with respect to current dataset

web_interface/static/js/presentation/right_panel/model/panelModelArchView.js

@@ -101,9 +101,9 @@ class PanelModelArchView extends PanelView {
         // Init all SVG primitives
         this.primitives = {}
 
-        // Function drawing an parameters data object
         let marginText = 5
         let marginBlocks = 140 // TODO make it = rightmost bound of all texts
+        // Function drawing an parameters data object
         let draw = (kv, primitives, offsets, keysList=[], depth=0) => {
             for (let [key, value] of Object.entries(kv)) {
                 let text
@@ -173,6 +173,15 @@ class PanelModelArchView extends PanelView {
                     primitives[key] = text
                     offsets[1] += 25
                 }
+                else if (typeof(value) === 'string') { // String, indicates tensor size is over limit
+                    this.svgPanel.$svg.append(Svg.text(
+                        `[Tensor ${value}]`,
+                        marginBlocks + 12*depth, offsets[1],
+                        'middle', '20px',
+                        'normal', "#000000"
+                    ))
+                    offsets[1] += this.size + 20
+                }
                 else if (value.constructor === Array) {
                     let arrayPrimitives = []
                     primitives[key] = arrayPrimitives
@@ -248,7 +257,7 @@ class PanelModelArchView extends PanelView {
                         }
                     }
                 }
-                else console.error("Unknown type")
+                else console.error("Model data contains unknown data type.")
             }
         }