Merge remote-tracking branch 'origin/main' into stateful

Author: slyalin

optimum/intel/openvino/modeling_base.py

@@ -30,7 +30,7 @@
 
 from ...exporters.openvino import export, main_export
 from ..utils.import_utils import is_nncf_available, is_transformers_version
-from .utils import ONNX_WEIGHTS_NAME, OV_XML_FILE_NAME
+from .utils import ONNX_WEIGHTS_NAME, OV_XML_FILE_NAME, _print_compiled_model_properties
 
 
 if is_transformers_version("<", "4.25.0"):
@@ -350,6 +350,10 @@ def compile(self):
                 ov_config["CACHE_DIR"] = str(cache_dir)
                 logger.info(f"Setting OpenVINO CACHE_DIR to {str(cache_dir)}")
             self.request = core.compile_model(self.model, self._device, ov_config)
+            # OPENVINO_LOG_LEVEL can be found in https://docs.openvino.ai/2023.2/openvino_docs_OV_UG_supported_plugins_AUTO_debugging.html
+            if "OPENVINO_LOG_LEVEL" in os.environ and int(os.environ["OPENVINO_LOG_LEVEL"]) > 2:
+                logger.info(f"{self._device} SUPPORTED_PROPERTIES:")
+                _print_compiled_model_properties(self.request)
 
     def _reshape(
         self,

optimum/intel/openvino/modeling_diffusion.py

@@ -55,7 +55,7 @@
 from ...exporters.openvino import main_export
 from .loaders import OVTextualInversionLoaderMixin
 from .modeling_base import OVBaseModel
-from .utils import ONNX_WEIGHTS_NAME, OV_TO_NP_TYPE, OV_XML_FILE_NAME
+from .utils import ONNX_WEIGHTS_NAME, OV_TO_NP_TYPE, OV_XML_FILE_NAME, _print_compiled_model_properties
 
 
 core = Core()
@@ -544,6 +544,10 @@ def _compile(self):
         if self.request is None:
             logger.info(f"Compiling the {self._model_name} to {self.device} ...")
             self.request = core.compile_model(self.model, self.device, self.ov_config)
+            # OPENVINO_LOG_LEVEL can be found in https://docs.openvino.ai/2023.2/openvino_docs_OV_UG_supported_plugins_AUTO_debugging.html
+            if "OPENVINO_LOG_LEVEL" in os.environ and int(os.environ["OPENVINO_LOG_LEVEL"]) > 2:
+                logger.info(f"{self.device} SUPPORTED_PROPERTIES:")
+                _print_compiled_model_properties(self.request)
 
     @property
     def device(self):

optimum/intel/openvino/modeling_seq2seq.py

@@ -14,6 +14,7 @@
 
 import copy
 import logging
+import os
 from pathlib import Path
 from tempfile import gettempdir
 from typing import TYPE_CHECKING, Dict, Optional, Tuple, Union
@@ -37,6 +38,7 @@
 
 from ..utils.import_utils import is_transformers_version
 from .modeling_base_seq2seq import OVBaseModelForSeq2SeqLM
+from .utils import _print_compiled_model_properties
 
 
 if is_transformers_version("<", "4.25.0"):
@@ -472,6 +474,10 @@ def _compile(self):
         if self.request is None:
             logger.info(f"Compiling the encoder to {self._device} ...")
             self.request = core.compile_model(self.model, self._device, self.ov_config)
+            # OPENVINO_LOG_LEVEL can be found in https://docs.openvino.ai/2023.2/openvino_docs_OV_UG_supported_plugins_AUTO_debugging.html
+            if "OPENVINO_LOG_LEVEL" in os.environ and int(os.environ["OPENVINO_LOG_LEVEL"]) > 2:
+                logger.info(f"{self._device} SUPPORTED_PROPERTIES:")
+                _print_compiled_model_properties(self.request)
 
 
 class OVDecoder:
@@ -570,7 +576,12 @@ def __call__(self, *args, **kwargs):
     def _compile(self):
         if self.request is None:
             logger.info(f"Compiling the decoder to {self._device} ...")
-            self.request = core.compile_model(self.model, self._device, self.ov_config).create_infer_request()
+            compiled_model = core.compile_model(self.model, self._device, self.ov_config)
+            self.request = compiled_model.create_infer_request()
+            # OPENVINO_LOG_LEVEL can be found in https://docs.openvino.ai/2023.2/openvino_docs_OV_UG_supported_plugins_AUTO_debugging.html
+            if "OPENVINO_LOG_LEVEL" in os.environ and int(os.environ["OPENVINO_LOG_LEVEL"]) > 2:
+                logger.info(f"{self._device} SUPPORTED_PROPERTIES:")
+                _print_compiled_model_properties(compiled_model)
 
 
 @add_start_docstrings(

optimum/intel/openvino/utils.py

@@ -14,15 +14,18 @@
 
 
 import json
+import logging
 import os
 from glob import glob
 
 import numpy as np
 from huggingface_hub import model_info
-from openvino.runtime import Type
+from openvino.runtime import Type, properties
 from transformers.onnx.utils import ParameterFormat, compute_serialized_parameters_size
 
 
+logger = logging.getLogger(__name__)
+
 OV_XML_FILE_NAME = "openvino_model.xml"
 OV_ENCODER_NAME = "openvino_encoder_model.xml"
 OV_DECODER_NAME = "openvino_decoder_model.xml"
@@ -123,3 +126,19 @@ def _is_timm_ov_dir(model_dir):
         if hf_hub_id and model_info(hf_hub_id).library_name == "timm":
             return True
     return False
+
+
+def _print_compiled_model_properties(compiled_model):
+    supported_properties = properties.supported_properties()
+    skip_keys = {"SUPPORTED_METRICS", "SUPPORTED_CONFIG_KEYS", supported_properties}
+    keys = set(compiled_model.get_property(supported_properties)) - skip_keys
+    for k in keys:
+        value = compiled_model.get_property(k)
+        if k == properties.device.properties():
+            for device_key in value.keys():
+                logger.info(f"  {device_key}:")
+                for k2, value2 in value.get(device_key).items():
+                    if k2 not in skip_keys:
+                        logger.info(f"    {k2}: {value2}")
+        else:
+            logger.info(f"  {k}: {value}")

setup.py

@@ -41,9 +41,9 @@
         "neural-compressor>=2.2.0",
         "onnx",
         "onnxruntime<1.15.0",
-        "transformers>=4.33.0",
+        "transformers>=4.34.0",
     ],
-    "openvino": ["openvino>=2023.2", "onnx", "onnxruntime", "transformers>=4.33.0"],
+    "openvino": ["openvino>=2023.2", "onnx", "onnxruntime", "transformers>=4.34.0"],
     "nncf": ["nncf>=2.7.0"],
     "ipex": ["transformers<4.32.0", "intel-extension-for-pytorch", "onnx"],
     "diffusers": ["diffusers"],