fix(onnx_loader): Uses default provider is providers are not supplied.

archai/discrete_search/evaluators/nlp/transformer_flex_latency.py

@@ -48,6 +48,10 @@ class TransformerFlexOnnxLatency(ModelEvaluator):
     ) -> None:
         """Initialize the evaluator.
 
+        This evaluator supports measuring in different ONNX Runtime providers. For measuring on
+        GPUs, use `providers=["CUDAExecutionProvider"]` and make sure that `onnxruntime-gpu`
+        package is installed.
+
         Args:
             search_space: The search space to use for loading the model.
             providers: The list of ORT providers to use for benchmarking.

archai/onnx/onnx_loader.py

@@ -37,6 +37,8 @@ def load_from_onnx(onnx_model_path: str, providers: Optional[List[str]] = None)
     options.intra_op_num_threads = OMP_NUM_THREADS
     options.graph_optimization_level = GraphOptimizationLevel.ORT_ENABLE_ALL
 
+    providers = providers or ["CPUExecutionProvider"]
+
     session = InferenceSession(onnx_model_path, sess_options=options, providers=providers)
     session.disable_fallback()
 

tasks/text_generation/search.py

@@ -107,7 +107,7 @@ if __name__ == "__main__":
     )
     search_objectives.add_objective(
         "onnx_latency",
-        TransformerFlexOnnxLatency(space, providers=["CPUExecutionProvider"], seq_len=1024, n_trials=5, use_past=False),
+        TransformerFlexOnnxLatency(space, seq_len=1024, n_trials=5, use_past=False),
         higher_is_better=False,
         compute_intensive=False,
     )