Merge branch 'main' into fix-main-export

optimum/commands/export/openvino.py

@@ -0,0 +1,106 @@
+# Copyright 2023 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+"""Defines the command line for the export with OpenVINO."""
+
+import sys
+from pathlib import Path
+from typing import TYPE_CHECKING, Optional
+
+from ...exporters import TasksManager
+from ..base import BaseOptimumCLICommand, CommandInfo
+
+
+if TYPE_CHECKING:
+    from argparse import ArgumentParser, Namespace, _SubParsersAction
+
+
+def parse_args_openvino(parser: "ArgumentParser"):
+    required_group = parser.add_argument_group("Required arguments")
+    required_group.add_argument(
+        "-m", "--model", type=str, required=True, help="Model ID on huggingface.co or path on disk to load model from."
+    )
+    required_group.add_argument(
+        "output", type=Path, help="Path indicating the directory where to store the generated OV model."
+    )
+    optional_group = parser.add_argument_group("Optional arguments")
+    optional_group.add_argument(
+        "--task",
+        default="auto",
+        help=(
+            "The task to export the model for. If not specified, the task will be auto-inferred based on the model. Available tasks depend on the model, but are among:"
+            f" {str(TasksManager.get_all_tasks())}. For decoder models, use `xxx-with-past` to export the model using past key values in the decoder."
+        ),
+    )
+    optional_group.add_argument("--cache_dir", type=str, default=None, help="Path indicating where to store cache.")
+    optional_group.add_argument(
+        "--framework",
+        type=str,
+        choices=["pt", "tf"],
+        default=None,
+        help=(
+            "The framework to use for the export. If not provided, will attempt to use the local checkpoint's original framework or what is available in the environment."
+        ),
+    )
+    optional_group.add_argument(
+        "--trust-remote-code",
+        action="store_true",
+        help=(
+            "Allows to use custom code for the modeling hosted in the model repository. This option should only be set for repositories you trust and in which "
+            "you have read the code, as it will execute on your local machine arbitrary code present in the model repository."
+        ),
+    )
+    optional_group.add_argument(
+        "--pad-token-id",
+        type=int,
+        default=None,
+        help=(
+            "This is needed by some models, for some tasks. If not provided, will attempt to use the tokenizer to guess it."
+        ),
+    )
+
+
+class OVExportCommand(BaseOptimumCLICommand):
+    COMMAND = CommandInfo(name="openvino", help="Export PyTorch models to OpenVINO IR.")
+
+    def __init__(
+        self,
+        subparsers: "_SubParsersAction",
+        args: Optional["Namespace"] = None,
+        command: Optional["CommandInfo"] = None,
+        from_defaults_factory: bool = False,
+        parser: Optional["ArgumentParser"] = None,
+    ):
+        super().__init__(
+            subparsers, args=args, command=command, from_defaults_factory=from_defaults_factory, parser=parser
+        )
+        self.args_string = " ".join(sys.argv[3:])
+
+    @staticmethod
+    def parse_args(parser: "ArgumentParser"):
+        return parse_args_openvino(parser)
+
+    def run(self):
+        from ...exporters.openvino.__main__ import main_export
+
+        # TODO : add input shapes
+        main_export(
+            model_name_or_path=self.args.model,
+            output=self.args.output,
+            task=self.args.task,
+            framework=self.args.framework,
+            cache_dir=self.args.cache_dir,
+            trust_remote_code=self.args.trust_remote_code,
+            pad_token_id=self.args.pad_token_id,
+            # **input_shapes,
+        )

optimum/commands/register/register_openvino.py

@@ -0,0 +1,19 @@
+# Copyright 2023 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from ..export import ExportCommand
+from ..export.openvino import OVExportCommand
+
+
+REGISTER_COMMANDS = [(OVExportCommand, ExportCommand)]

tests/openvino/test_exporters_cli.py

@@ -0,0 +1,59 @@
+# Copyright 2023 The HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+import subprocess
+import unittest
+from tempfile import TemporaryDirectory
+
+from parameterized import parameterized
+from utils_tests import MODEL_NAMES
+
+from optimum.exporters.openvino.__main__ import main_export
+
+
+class OVCLIExportTestCase(unittest.TestCase):
+    """
+    Integration tests ensuring supported models are correctly exported.
+    """
+
+    SUPPORTED_ARCHITECTURES = (
+        ["causal-lm", "gpt2"],
+        ["causal-lm-with-past", "gpt2"],
+        ["seq2seq-lm", "t5"],
+        ["seq2seq-lm-with-past", "t5"],
+        ["sequence-classification", "bert"],
+        ["question-answering", "distilbert"],
+        ["masked-lm", "bert"],
+        ["default", "blenderbot"],
+        ["default-with-past", "blenderbot"],
+        ["stable-diffusion", "stable-diffusion"],
+        ["stable-diffusion-xl", "stable-diffusion-xl"],
+        ["stable-diffusion-xl", "stable-diffusion-xl-refiner"],
+    )
+
+    def _openvino_export(self, model_name: str, task: str):
+        with TemporaryDirectory() as tmpdir:
+            main_export(model_name_or_path=model_name, output=tmpdir, task=task)
+
+    @parameterized.expand(SUPPORTED_ARCHITECTURES)
+    def test_export(self, task: str, model_type: str):
+        self._openvino_export(MODEL_NAMES[model_type], task)
+
+    @parameterized.expand(SUPPORTED_ARCHITECTURES)
+    def test_exporters_cli(self, task: str, model_type: str):
+        with TemporaryDirectory() as tmpdirname:
+            subprocess.run(
+                f"optimum-cli export openvino --model {MODEL_NAMES[model_type]} --task {task} {tmpdirname}",
+                shell=True,
+                check=True,
+            )

tests/openvino/test_modeling.py

@@ -121,10 +121,12 @@ def test_load_from_hub_and_save_model(self):
         del model
         gc.collect()
 
-    def test_load_from_hub_and_save_decoder_model(self):
-        tokenizer = AutoTokenizer.from_pretrained(self.OV_DECODER_MODEL_ID)
+    @parameterized.expand((True, False))
+    def test_load_from_hub_and_save_decoder_model(self, use_cache):
+        model_id = "vuiseng9/ov-gpt2-fp32-kv-cache" if use_cache else "vuiseng9/ov-gpt2-fp32-no-cache"
+        tokenizer = AutoTokenizer.from_pretrained(model_id)
         tokens = tokenizer("This is a sample input", return_tensors="pt")
-        loaded_model = OVModelForCausalLM.from_pretrained(self.OV_DECODER_MODEL_ID, use_cache=True)
+        loaded_model = OVModelForCausalLM.from_pretrained(model_id, use_cache=use_cache)
         self.assertIsInstance(loaded_model.config, PretrainedConfig)
         loaded_model_outputs = loaded_model(**tokens)
 
@@ -133,7 +135,8 @@ def test_load_from_hub_and_save_decoder_model(self):
             folder_contents = os.listdir(tmpdirname)
             self.assertTrue(OV_XML_FILE_NAME in folder_contents)
             self.assertTrue(OV_XML_FILE_NAME.replace(".xml", ".bin") in folder_contents)
-            model = OVModelForCausalLM.from_pretrained(tmpdirname, use_cache=True)
+            model = OVModelForCausalLM.from_pretrained(tmpdirname, use_cache=use_cache)
+            self.assertEqual(model.use_cache, use_cache)
 
         outputs = model(**tokens)
         self.assertTrue(torch.equal(loaded_model_outputs.logits, outputs.logits))
@@ -540,6 +543,7 @@ def test_compare_with_and_without_past_key_values(self):
             )
 
         model_without_pkv = OVModelForCausalLM.from_pretrained(model_id, export=True, use_cache=False)
+
         # Warmup
         _ = model_without_pkv.generate(**tokens)
         with Timer() as without_pkv_timer:
@@ -710,12 +714,8 @@ def test_timm_save_and_infer(self, model_id):
         with tempfile.TemporaryDirectory() as tmpdirname:
             model_save_path = os.path.join(tmpdirname, "timm_ov_model")
             ov_model.save_pretrained(model_save_path)
-            new_ov_model = OVModelForImageClassification.from_pretrained(
-                model_save_path,
-            )
-            new_ov_model(
-                pixel_values=torch.zeros((5, 3, new_ov_model.config.image_size, new_ov_model.config.image_size))
-            )
+            model = OVModelForImageClassification.from_pretrained(model_save_path)
+            model(pixel_values=torch.zeros((5, 3, model.config.image_size, model.config.image_size)))
         gc.collect()