Fix default ov config (#600)

Author: echarlaix

optimum/intel/openvino/modeling.py

@@ -434,8 +434,8 @@ def _from_transformers(
         save_dir = TemporaryDirectory()
         save_dir_path = Path(save_dir.name)
 
-        # If load_in_8bit or quantization_config not specified then ov_config is set to None and will be set by default in convert depending on the model size
-        if load_in_8bit is None or not quantization_config:
+        # If load_in_8bit and quantization_config not specified then ov_config is set to None and will be set by default in convert depending on the model size
+        if load_in_8bit is None and not quantization_config:
             ov_config = None
         else:
             ov_config = OVConfig(dtype="fp32")

optimum/intel/openvino/modeling_base.py

@@ -314,8 +314,8 @@ def _from_transformers(
         save_dir = TemporaryDirectory()
         save_dir_path = Path(save_dir.name)
 
-        # If load_in_8bit or quantization_config not specified then ov_config is set to None and will be set by default in convert depending on the model size
-        if load_in_8bit is None or not quantization_config:
+        # If load_in_8bit and quantization_config not specified then ov_config is set to None and will be set by default in convert depending on the model size
+        if load_in_8bit is None and not quantization_config:
             ov_config = None
         else:
             ov_config = OVConfig(dtype="fp32")

optimum/intel/openvino/modeling_base_seq2seq.py

@@ -258,8 +258,8 @@ def _from_transformers(
             if use_cache:
                 task = task + "-with-past"
 
-        # If load_in_8bit or quantization_config not specified then ov_config is set to None and will be set by default in convert depending on the model size
-        if load_in_8bit is None or not quantization_config:
+        # If load_in_8bit and quantization_config not specified then ov_config is set to None and will be set by default in convert depending on the model size
+        if load_in_8bit is None and not quantization_config:
             ov_config = None
         else:
             ov_config = OVConfig(dtype="fp32")

optimum/intel/openvino/modeling_decoder.py

@@ -260,7 +260,7 @@ def _from_transformers(
             if use_cache:
                 task = task + "-with-past"
 
-        # If load_in_8bit or quantization_config not specified then ov_config is set to None and will be set by default in convert depending on the model size
+        # If load_in_8bit and quantization_config not specified then ov_config is set to None and will be set by default in convert depending on the model size
         if load_in_8bit is None and not quantization_config:
             ov_export_config = None
         else:

optimum/intel/openvino/modeling_diffusion.py

@@ -321,8 +321,8 @@ def _from_transformers(
         save_dir = TemporaryDirectory()
         save_dir_path = Path(save_dir.name)
 
-        # If load_in_8bit or quantization_config not specified then ov_config is set to None and will be set by default in convert depending on the model size
-        if load_in_8bit is None or not quantization_config:
+        # If load_in_8bit and quantization_config not specified then ov_config is set to None and will be set by default in convert depending on the model size
+        if load_in_8bit is None and not quantization_config:
             ov_config = None
         else:
             ov_config = OVConfig(dtype="fp32")