Divide iree_utils and do module imports on function calls.

Author: Prashant Kumar

.github/workflows/nightly.yml

@@ -53,15 +53,14 @@ jobs:
     - name: Install dependencies
       run: |
         python -m pip install --upgrade pip
-        python -m pip install flake8 pytest yapf toml
+        python -m pip install flake8 pytest toml
         if [ -f requirements.txt ]; then pip install -r requirements.txt --extra-index-url https://download.pytorch.org/whl/nightly/cpu  -f https://github.com/llvm/torch-mlir/releases -f https://github.com/nod-ai/SHARK-Runtime/releases; fi
     - name: Lint with flake8
       run: |
         # stop the build if there are Python syntax errors or undefined names
         flake8 . --count --select=E9,F63,F7,F82 --show-source --statistics --exclude shark.venv,lit.cfg.py 
         # exit-zero treats all errors as warnings. The GitHub editor is 127 chars wide
         flake8 . --count --exit-zero --max-complexity=10 --max-line-length=127 --statistics --exclude shark.venv,lit.cfg.py 
-        yapf -i --style .style.yapf shark/*.py
 
     - name: Build and validate the package
       run: |

.github/workflows/python-format.yml

@@ -0,0 +1,15 @@
+name: black-formatter
+on: [pull_request]
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v1
+      - name: Set up Python 3.10
+        uses: actions/setup-python@v1
+        with:
+          python-version: 3.10.5
+      - name: Install Black
+        run: pip install black
+      - name: Run formatter check on the entire project.
+        run: black --line-length 80 --check .

.github/workflows/test-models.yml

@@ -37,15 +37,14 @@ jobs:
     - name: Install dependencies
       run: |
         python -m pip install --upgrade pip
-        python -m pip install flake8 pytest yapf toml
+        python -m pip install flake8 pytest toml
 
     - name: Lint with flake8
       run: |
         # stop the build if there are Python syntax errors or undefined names
         flake8 . --count --select=E9,F63,F7,F82 --show-source --statistics --exclude lit.cfg.py
         # exit-zero treats all errors as warnings. The GitHub editor is 127 chars wide
         flake8 . --count --exit-zero --max-complexity=10 --max-line-length=127 --statistics --exclude lit.cfg.py
-        yapf -i --style .style.yapf shark/*.py
 
     - name: Validate Models
       run: |

benchmarks/hf_model_benchmark.py

@@ -6,16 +6,16 @@
     "--model_name",
     type=str,
     required=True,
-    help=
-    "Specifies name of HF model to benchmark. (For exmaple \"microsoft/MiniLM-L12-H384-uncased\""
+    help='Specifies name of HF model to benchmark. (For exmaple "microsoft/MiniLM-L12-H384-uncased"',
 )
 load_args, unknown = parser.parse_known_args()
 
 if __name__ == "__main__":
     model_name = load_args.model_name
     test_input = torch.randint(2, (1, 128))
-    shark_module = SharkHFBenchmarkRunner(model_name, (test_input,),
-                                          jit_trace=True)
+    shark_module = SharkHFBenchmarkRunner(
+        model_name, (test_input,), jit_trace=True
+    )
     shark_module.benchmark_c()
     shark_module.benchmark_python((test_input,))
     shark_module.benchmark_torch(test_input)

benchmarks/hf_transformer.py

@@ -1,16 +1,19 @@
 import torch
-from shark.shark_runner import SharkBenchmarkRunner
+from shark.shark_benchmark_runner import SharkBenchmarkRunner
 from shark.parser import shark_args
 from transformers import AutoTokenizer, AutoModelForSequenceClassification
-from onnxruntime.transformers.benchmark import run_pytorch, run_tensorflow, run_onnxruntime
+from onnxruntime.transformers.benchmark import (
+    run_pytorch,
+    run_tensorflow,
+    run_onnxruntime,
+)
 from onnxruntime.transformers.huggingface_models import MODELS
 from onnxruntime.transformers.benchmark_helper import ConfigModifier, Precision
 import os
 import psutil
 
 
 class OnnxFusionOptions(object):
-
     def __init__(self):
         self.disable_gelu = False
         self.disable_layer_norm = False
@@ -25,17 +28,13 @@ def __init__(self):
 
 
 class HuggingFaceLanguage(torch.nn.Module):
-
     def __init__(self, hf_model_name):
         super().__init__()
         self.model = AutoModelForSequenceClassification.from_pretrained(
             hf_model_name,  # The pretrained model.
-            num_labels=
-            2,  # The number of output labels--2 for binary classification.
-            output_attentions=
-            False,  # Whether the model returns attentions weights.
-            output_hidden_states=
-            False,  # Whether the model returns all hidden-states.
+            num_labels=2,  # The number of output labels--2 for binary classification.
+            output_attentions=False,  # Whether the model returns attentions weights.
+            output_hidden_states=False,  # Whether the model returns all hidden-states.
             torchscript=True,
         )
 
@@ -62,8 +61,16 @@ def __init__(
             )
         self.model_name = model_name
         model = HuggingFaceLanguage(model_name)
-        SharkBenchmarkRunner.__init__(self, model, input, dynamic, self.device,
-                                      jit_trace, from_aot, frontend)
+        SharkBenchmarkRunner.__init__(
+            self,
+            model,
+            input,
+            dynamic,
+            self.device,
+            jit_trace,
+            from_aot,
+            frontend,
+        )
 
     def benchmark_torch(self, inputs):
         use_gpu = self.device == "gpu"
@@ -74,10 +81,20 @@ def benchmark_torch(self, inputs):
         sequence_lengths = [inputs.shape[-1]]
         cache_dir = os.path.join(".", "cache_models")
         verbose = False
-        result = run_pytorch(use_gpu, [self.model_name], None, config_modifier,
-                             Precision.FLOAT32, num_threads, batch_sizes,
-                             sequence_lengths, shark_args.num_iterations, False,
-                             cache_dir, verbose)
+        result = run_pytorch(
+            use_gpu,
+            [self.model_name],
+            None,
+            config_modifier,
+            Precision.FLOAT32,
+            num_threads,
+            batch_sizes,
+            sequence_lengths,
+            shark_args.num_iterations,
+            False,
+            cache_dir,
+            verbose,
+        )
         print(
             f"ONNX Pytorch-benchmark:{result[0]['QPS']} iter/second, Total Iterations:{shark_args.num_iterations}"
         )
@@ -92,10 +109,19 @@ def benchmark_tf(self, inputs):
         sequence_lengths = [inputs.shape[-1]]
         cache_dir = os.path.join(".", "cache_models")
         verbose = False
-        result = run_tensorflow(use_gpu, [self.model_name], None,
-                                config_modifier, Precision.FLOAT32, num_threads,
-                                batch_sizes, sequence_lengths,
-                                shark_args.num_iterations, cache_dir, verbose)
+        result = run_tensorflow(
+            use_gpu,
+            [self.model_name],
+            None,
+            config_modifier,
+            Precision.FLOAT32,
+            num_threads,
+            batch_sizes,
+            sequence_lengths,
+            shark_args.num_iterations,
+            cache_dir,
+            verbose,
+        )
         print(
             f"ONNX TF-benchmark:{result[0]['QPS']} iter/second, Total Iterations:{shark_args.num_iterations}"
         )
@@ -105,7 +131,8 @@ def benchmark_onnx(self, inputs):
             print(
                 f"{self.model_name} is currently not supported in ORT's HF. Check \
 https://github.com/microsoft/onnxruntime/blob/master/onnxruntime/python/tools/transformers/huggingface_models.py \
-for currently supported models. Exiting benchmark ONNX.")
+for currently supported models. Exiting benchmark ONNX."
+            )
             return
         use_gpu = self.device == "gpu"
         num_threads = psutil.cpu_count(logical=False)
@@ -121,17 +148,34 @@ def benchmark_onnx(self, inputs):
         use_raw_attention_mask = True
         model_fusion_statistics = {}
         overwrite = False
-        model_source = "pt"  #Either "pt" or "tf"
+        model_source = "pt"  # Either "pt" or "tf"
         provider = None
         config_modifier = ConfigModifier(None)
         onnx_args = OnnxFusionOptions()
         result = run_onnxruntime(
-            use_gpu, provider, [self.model_name], None, config_modifier,
-            Precision.FLOAT32, num_threads, batch_sizes, sequence_lengths,
-            shark_args.num_iterations, input_counts, optimize_onnx,
-            validate_onnx, cache_dir, onnx_dir, verbose, overwrite,
-            disable_ort_io_binding, use_raw_attention_mask,
-            model_fusion_statistics, model_source, onnx_args)
+            use_gpu,
+            provider,
+            [self.model_name],
+            None,
+            config_modifier,
+            Precision.FLOAT32,
+            num_threads,
+            batch_sizes,
+            sequence_lengths,
+            shark_args.num_iterations,
+            input_counts,
+            optimize_onnx,
+            validate_onnx,
+            cache_dir,
+            onnx_dir,
+            verbose,
+            overwrite,
+            disable_ort_io_binding,
+            use_raw_attention_mask,
+            model_fusion_statistics,
+            model_source,
+            onnx_args,
+        )
         print(
             f"ONNX ORT-benchmark:{result[0]['QPS']} iter/second, Total Iterations:{shark_args.num_iterations}"
         )