Truncated gpt model examples for iree debug

alt_e2eshark/e2e_testing/storage.py

@@ -142,7 +142,9 @@ class TestTensors:
 
     def __init__(self, data: Tuple):
         self.data = data
-        self.type = type(self.data[0])
+        self.type = None
+        if len(data) > 0 :
+            self.type = type(self.data[0])
         if not all([type(d) == self.type for d in data]):
             self.type == None
 

alt_e2eshark/onnx_tests/helper_classes.py

@@ -75,6 +75,7 @@ def __init__(self, og_model_info_class: type, og_name: str, *args, **kwargs):
         run_dir = Path(self.model).parents[1]
         og_model_path = os.path.join(run_dir, og_name)
         self.sibling_inst = og_model_info_class(og_name, og_model_path)
+        self.opset_version = self.sibling_inst.opset_version
 
     def construct_model(self):
         if not os.path.exists(self.sibling_inst.model):

alt_e2eshark/onnx_tests/models/nlp.py

@@ -100,3 +100,43 @@ def construct_inputs(self):
         register_test(dim_param_constructor(dim_params), model_name)
 
 # You can add more customizations or specific handling for certain models here
+from ..helper_classes import TruncatedModel, get_trucated_constructor
+# download model.onnx to /proj/gdba/shark/chi/src/SHARK-TestSuite/alt_e2eshark/tmp/mygpt4
+# export CACHE_DIR=/proj/gdba/shark/chi/src/SHARK-TestSuite/alt_e2eshark/tmp
+model_name = "mygpt4"
+mygpt4_nlp_params = {
+    "batch_size": 1,
+    "seq_len": 128,
+    "encoder_sequence_length": 128,
+    "decoder_sequence_length": 128,
+    "unk__2829": 1,
+    "unk__2828": 1,
+    "unk__2827": 1,
+    "unk__2826": 1,
+    "unk__2825": 1,
+    "unk__2824": 1,
+}
+t_model_constructor = get_trucated_constructor(TruncatedModel, dim_param_constructor(mygpt4_nlp_params), model_name)
+# Run to the last onnx ops / the whole model with:
+# python ./run.py --mode=cl-onnx-iree -v -t mygpt4
+#   Stages to be run: ['setup', 'import_model', 'preprocessing', 'compilation', 'construct_inputs', 'native_inference', 'compiled_inference', 'postprocessing']
+#   Test list: ['mygpt4']
+#   running test mygpt4...
+#   Unzipping - /proj/gdba/shark/chi/src/SHARK-TestSuite/alt_e2eshark/tmp/mygpt4/model.onnx.zip... 
+#   Unzipping succeded. Look for extracted contents in /proj/gdba/shark/chi/src/SHARK-TestSuite/alt_e2eshark/test-run/mygpt4
+#   {'Shape': 63, 'Cast': 141, 'Slice': 136, 'Squeeze': 25, 'Range': 25, 'Concat': 65, 'Reshape': 161, 'Unsqueeze': 38, 'Sub': 85, 'Mul': 169, 'Add': 159, 'Gather': 3, 'MatMul': 72, 'Split': 12, 'Transpose': 48, 'GreaterOrEqual': 12, 'Softmax': 12, 'GlobalAveragePool': 48, 'Sqrt': 24, 'Reciprocal': 24, 'Erf': 12}#         
+#   Running stage 'import_model'...      
+register_test(t_model_constructor(1, ""), "mygpt4")
+# Run to the last 2/7/12/17 onnx op with:
+# run with python ./run.py --mode=cl-onnx-iree -v -t mygpt4_trunc_
+# for n in range(2, 20, 5):
+#     register_test(t_model_constructor(n,""), f"mygpt4_trunc_{n}")
+
+# Run to 5/55/105/155 onnx.Add ops with:
+# run with python ./run.py --mode=cl-onnx-iree -v -t mygpt4_trunc_add_
+for n in range(5, 160, 50):
+    register_test(t_model_constructor(n,"Add"), f"mygpt4_trunc_add_{n}")
+
+# python ./run.py --mode=cl-onnx-iree -v -t mygpt4_trunc_shape_1
+register_test(t_model_constructor(1, "Shape"), "mygpt4_trunc_shape_1")
+register_test(t_model_constructor(161, "Reshape"), "mygpt4_trunc_reshape_161")