Add CV/NLP SymDim Reifier matching error msg. (#502)

* Add CV/NLP SymDim Reifier matching error msg.

* Add missing reifier of model.

Author: TelGome

graph_net/torch/sym_dim_reifiers/naive_cv_sym_dim_reifier.py

@@ -16,7 +16,14 @@ def match(self) -> bool:
         if self.dyn_dim_cstrs is None:
             return False
         sym_shapes_str = self.dyn_dim_cstrs.serialize_symbolic_input_shapes_to_str()
-        return sym_shapes_str in self._get_map_cv_sym_shapes_str2reifier()
+        if sym_shapes_str == "[]":
+            return False
+        elif sym_shapes_str not in self._get_map_cv_sym_shapes_str2reifier():
+            print(
+                f"[CV SymDim Reify] No reifier matched symbolic shapes:{sym_shapes_str} \nPlease add a reify rule to _get_map_cv_sym_shapes_str2reifier()"
+            )
+            return False
+        return True
 
     def reify(self):
         assert self.match()
@@ -40,6 +47,15 @@ def _get_map_cv_sym_shapes_str2reifier(cls):
                 "[(S0,80,S1),(S0,)]": cls.reify_nemo_asr_s0_s1,
                 "[(S0,3,512,1024)]": cls.reify_semantic_seg_s0,
                 "[(S0,3,640,640)]": cls.reify_yolo_s0,
+                "[(S0,)]": cls.reify_s0,
+                "[(S0,S1,S2)]": cls.reify_s0_s1_s2,
+                "[(S0,S1,S2,S2)]": cls.reify_s0_s1_s2,
+                "[(S0,3,S1,S2,S2)]": cls.reify_s0_s1_s2,
+                "[(S0,S1,S1,384)]": cls.reify_s0_s1_384,
+                "[(S0,S1),(S0,S1,2560)]": cls.reify_s0_s1_2560,
+                "[(S0,S1,256),(S0,S1,256)]": cls.reify_s0_s1_256,
+                "[(S0,S1),(S0,S1,3072)]": cls.reify_s0_s1_128_1024,
+                "[(S0,S1),(S0,S1,1024),(S0,S1,1024)]": cls.reify_s0_s1_128_1024,
             }
         return cls.g_cv_sym_shapes_str2reifier
 
@@ -184,3 +200,89 @@ def reify_yolo_s0(self):
         return {
             S0: [[1], [2], [4], [8], [12], [16], [24], [32], [64]],
         }
+
+    def reify_s0(self):
+        S0 = (sympy.Symbol("S0"),)
+        return {
+            S0: [[1], [2], [4], [8], [16], [32], [48], [64], [128]],
+        }
+
+    def reify_s0_s1_s2(self):
+        S0S1S2 = (sympy.Symbol("S0"), sympy.Symbol("S1"), sympy.Symbol("S2"))
+        return {
+            S0S1S2: [
+                [1, 8, 8],
+                [1, 8, 16],
+                [1, 16, 16],
+                [2, 16, 16],
+                [1, 16, 32],
+                [2, 16, 32],
+                [4, 16, 32],
+                [1, 32, 32],
+                [2, 32, 32],
+            ],
+        }
+
+    def reify_s0_s1_384(self):
+        S0S1 = (sympy.Symbol("S0"), sympy.Symbol("S1"))
+        return {
+            S0S1: [
+                [1, 64],
+                [1, 128],
+                [1, 192],
+                [1, 224],
+                [1, 256],
+                [4, 224],
+                [8, 224],
+                [32, 224],
+                [64, 224],
+            ],
+        }
+
+    def reify_s0_s1_2560(self):
+        S0S1 = (sympy.Symbol("S0"), sympy.Symbol("S1"))
+        return {
+            S0S1: [
+                [1, 16],
+                [1, 32],
+                [2, 32],
+                [1, 64],
+                [2, 64],
+                [1, 128],
+                [2, 128],
+                [1, 256],
+                [2, 256],
+            ],
+        }
+
+    def reify_s0_s1_256(self):
+        S0S1 = (sympy.Symbol("S0"), sympy.Symbol("S1"))
+        return {
+            S0S1: [
+                [1, 8],
+                [1, 16],
+                [2, 16],
+                [1, 32],
+                [2, 32],
+                [4, 32],
+                [1, 64],
+                [2, 64],
+                [4, 64],
+            ],
+        }
+
+    def reify_s0_s1_128_1024(self):
+        S0S1 = (sympy.Symbol("S0"), sympy.Symbol("S1"))
+        return {
+            S0S1: [
+                [1, 8],
+                [1, 16],
+                [2, 16],
+                [1, 32],
+                [2, 32],
+                [4, 32],
+                [1, 64],
+                [2, 64],
+                [4, 64],
+            ],
+        }

graph_net/torch/sym_dim_reifiers/naive_nlp_sym_dim_reifier.py

@@ -16,7 +16,14 @@ def match(self) -> bool:
         if self.dyn_dim_cstrs is None:
             return False
         sym_shapes_str = self.dyn_dim_cstrs.serialize_symbolic_input_shapes_to_str()
-        return sym_shapes_str in self._get_map_nlp_sym_shapes_str2reifier()
+        if sym_shapes_str == "[]":
+            return False
+        elif sym_shapes_str not in self._get_map_nlp_sym_shapes_str2reifier():
+            print(
+                f"[NLP SymDim Reify] No reifier matched symbolic shapes:{sym_shapes_str} \nPlease add a reify rule to _get_map_nlp_sym_shapes_str2reifier()"
+            )
+            return False
+        return True
 
     def reify(self):
         assert self.match()

samples/timm/sequencer2d_l.in1k/graph_net.json

@@ -4,5 +4,6 @@
     "num_nodes_required": 1,
     "dynamic": true,
     "source": "timm",
-    "heuristic_tag": "computer_vision"
+    "heuristic_tag": "computer_vision",
+    "symbolic_dimension_reifier": "naive_cv_sym_dim_reifier"
 }

samples/torchgeometric/MetaPath2Vec/graph_net.json

@@ -1 +1,10 @@
-{"framework": "torch", "num_devices_required": 1, "num_nodes_required": 1, "dynamic": false, "source": "torchgeometric", "heuristic_tag": "other", "dimension_generalization_passes": []}
+{
+    "framework": "torch",
+    "num_devices_required": 1,
+    "num_nodes_required": 1,
+    "dynamic": false,
+    "source": "torchgeometric",
+    "heuristic_tag": "other",
+    "dimension_generalization_passes": [],
+    "symbolic_dimension_reifier": "naive_cv_sym_dim_reifier"
+}

samples/torchgeometric/Node2Vec/graph_net.json

@@ -1 +1,10 @@
-{"framework": "torch", "num_devices_required": 1, "num_nodes_required": 1, "dynamic": false, "source": "torchgeometric", "heuristic_tag": "other", "dimension_generalization_passes": []}
+{
+    "framework": "torch",
+    "num_devices_required": 1,
+    "num_nodes_required": 1,
+    "dynamic": false,
+    "source": "torchgeometric",
+    "heuristic_tag": "other",
+    "dimension_generalization_passes": [],
+    "symbolic_dimension_reifier": "naive_cv_sym_dim_reifier"
+}

samples/torchvision/googlenet/graph_net.json

@@ -1 +1,9 @@
-{"framework": "torch", "num_devices_required": 1, "num_nodes_required": 1, "source": "torchvision", "heuristic_tag": "computer_vision", "dimension_generalization_passes": []}
+{
+    "framework": "torch",
+    "num_devices_required": 1,
+    "num_nodes_required": 1,
+    "source": "torchvision",
+    "heuristic_tag": "computer_vision",
+    "dimension_generalization_passes": [],
+    "symbolic_dimension_reifier": "naive_cv_sym_dim_reifier"
+}

samples/torchvision/inception_v3/graph_net.json

@@ -1 +1,9 @@
-{"framework": "torch", "num_devices_required": 1, "num_nodes_required": 1, "source": "torchvision", "heuristic_tag": "computer_vision", "dimension_generalization_passes": []}
+{
+    "framework": "torch",
+    "num_devices_required": 1,
+    "num_nodes_required": 1,
+    "source": "torchvision",
+    "heuristic_tag": "computer_vision",
+    "dimension_generalization_passes": [],
+    "symbolic_dimension_reifier": "naive_cv_sym_dim_reifier"
+}

samples/torchvision/r3d_18/graph_net.json

@@ -1 +1,9 @@
-{"framework": "torch", "num_devices_required": 1, "num_nodes_required": 1, "source": "torchvision", "heuristic_tag": "computer_vision", "dimension_generalization_passes": []}
+{
+    "framework": "torch",
+    "num_devices_required": 1,
+    "num_nodes_required": 1,
+    "source": "torchvision",
+    "heuristic_tag": "computer_vision",
+    "dimension_generalization_passes": [],
+    "symbolic_dimension_reifier": "naive_cv_sym_dim_reifier"
+}

samples/transformers-auto-model/IDEA-Research_grounding-dino-tiny/graph_net.json

@@ -16,5 +16,6 @@
         "license:apache-2.0",
         "region:us"
     ],
-    "heuristic_tag": "other"
+    "heuristic_tag": "other",
+    "symbolic_dimension_reifier": "naive_cv_sym_dim_reifier"
 }

samples/transformers-auto-model/UAE-Large-V1/graph_net.json

@@ -26,5 +26,6 @@
         "endpoints_compatible",
         "region:us"
     ],
-    "heuristic_tag": "other"
+    "heuristic_tag": "other",
+    "symbolic_dimension_reifier": "naive_cv_sym_dim_reifier"
 }