fix ipex

Author: IlyasMoutawwakil

examples/cpu_ipex_bert.yaml

@@ -16,7 +16,6 @@ launcher:
 
 backend:
   device: cpu
-  export: true
   no_weights: false # on multi-node machines, inline weights initialization harms performance
   torch_dtype: float32 # use bfloat16 on compatible Intel CPUs
   model: google-bert/bert-base-uncased

examples/cpu_ipex_llama.yaml

@@ -16,7 +16,6 @@ launcher:
 
 backend:
   device: cpu
-  export: true
   no_weights: false # on multi-node machines, intializing weights in the benchmark could harm performance
   torch_dtype: float32 # use bfloat16 on compatible Intel CPUs
   model: TinyLlama/TinyLlama-1.1B-Chat-v1.0

optimum_benchmark/backends/ipex/backend.py

@@ -55,18 +55,13 @@ def load_ipexmodel_from_pretrained(self) -> None:
     def load_ipexmodel_with_no_weights(self) -> None:
         with fast_weights_init():
             original_model, self.config.model = self.config.model, self.no_weights_model_path.as_posix()
-            original_export, self.config.export = self.config.export, True
             self.load_ipexmodel_from_pretrained()
-            self.config.export = original_export
             self.config.model = original_model
 
     @property
     def ipexmodel_kwargs(self) -> Dict[str, Any]:
         kwargs = {}
 
-        if self.config.export:
-            kwargs["export"] = self.config.export
-
         if self.config.torch_dtype is not None:
             kwargs["torch_dtype"] = getattr(torch, self.config.torch_dtype)
 

optimum_benchmark/backends/ipex/config.py

@@ -16,7 +16,6 @@ class IPEXConfig(BackendConfig):
     no_weights: bool = False
 
     # ipexmodel kwargs
-    export: Optional[bool] = None
     torch_dtype: Optional[str] = None
 
     def __post_init__(self):

tests/test_examples.py

@@ -59,7 +59,7 @@ def test_cli_configs(config_name):
         "--config-name",
         config_name,
         "scenario.warmup_runs=1",
-        "scenario.input_shapes.batch_size=2",
+        "scenario.input_shapes.batch_size=1",
         "++scenario.input_shapes.sequence_length=16",
         "++scenario.generate_kwargs.max_new_tokens=16",
         "++scenario.generate_kwargs.min_new_tokens=16",