Add support for nomic-ai/nomic-embed-text-v1.5 model

optimum/commands/env.py

@@ -57,8 +57,8 @@ def run(self):
             "Platform": platform.platform(),
             "Python version": platform.python_version(),
             "Huggingface_hub version": huggingface_hub.__version__,
-            "PyTorch version (GPU?)": f"{pt_version} (cuda availabe: {pt_cuda_available})",
-            "Tensorflow version (GPU?)": f"{tf_version} (cuda availabe: {tf_cuda_available})",
+            "PyTorch version (GPU?)": f"{pt_version} (cuda available: {pt_cuda_available})",
+            "Tensorflow version (GPU?)": f"{tf_version} (cuda available: {tf_cuda_available})",
         }
 
         print("\nCopy-and-paste the text below in your GitHub issue:\n")

optimum/exporters/onnx/base.py

@@ -373,7 +373,7 @@ def is_transformers_support_available(self) -> bool:
         Whether the installed version of Transformers allows for the ONNX export.
 
         Returns:
-            `bool`: Whether the install version of Transformers is compatible with the model.
+            `bool`: Whether the installed version of Transformers is compatible with the model.
 
         """
         return check_if_transformers_greater(self.MIN_TRANSFORMERS_VERSION)

optimum/exporters/onnx/model_configs.py

@@ -115,6 +115,10 @@ def inputs(self) -> Dict[str, Dict[int, str]]:
         }
 
 
+class NomicBertOnnxConfig(BertOnnxConfig):
+    DEFAULT_ONNX_OPSET = 14
+
+
 class AlbertOnnxConfig(BertOnnxConfig):
     DEFAULT_ONNX_OPSET = 11
 

optimum/exporters/tasks.py

@@ -828,6 +828,7 @@ class TasksManager:
             "text2text-generation-with-past",
             onnx="M2M100OnnxConfig",
         ),
+        "nomic-bert": supported_tasks_mapping("feature-extraction", onnx="NomicBertOnnxConfig"),
         "nystromformer": supported_tasks_mapping(
             "feature-extraction",
             "fill-mask",
@@ -1152,7 +1153,7 @@ def create_register(
             backend (`str`):
                 The name of the backend that the register function will handle.
             overwrite_existing (`bool`, defaults to `False`):
-                Whether or not the register function is allowed to overwrite an already existing config.
+                Whether the register function is allowed to overwrite an already existing config.
 
         Returns:
             `Callable[[str, Tuple[str, ...]], Callable[[Type], Type]]`: A decorator taking the model type and a the

optimum/utils/normalized_config.py

@@ -259,6 +259,7 @@ class NormalizedConfigManager:
         "mpt": MPTNormalizedTextConfig,
         "mt5": T5LikeNormalizedTextConfig,
         "m2m-100": BartLikeNormalizedTextConfig,
+        "nomic-bert": NormalizedTextConfig,
         "nystromformer": NormalizedTextConfig,
         "opt": NormalizedTextConfig,
         "pegasus": BartLikeNormalizedTextConfig,

tests/exporters/exporters_utils.py

@@ -299,6 +299,10 @@
     "stable-diffusion-xl": "echarlaix/tiny-random-stable-diffusion-xl",
 }
 
+PYTORCH_REMOTE_CODE_MODELS = {
+    "nomic-bert": "nomic-ai/nomic-embed-text-v1.5"
+}
+
 PYTORCH_TIMM_MODEL = {
     "default-timm-config": {
         "timm/inception_v3.tf_adv_in1k": ["image-classification"],

tests/exporters/onnx/test_exporters_onnx_cli.py

@@ -47,6 +47,7 @@
     PYTORCH_TIMM_MODEL,
     PYTORCH_TIMM_MODEL_NO_DYNAMIC_AXES,
     PYTORCH_TRANSFORMERS_MODEL_NO_DYNAMIC_AXES,
+    PYTORCH_REMOTE_CODE_MODELS
 )
 
 
@@ -730,3 +731,32 @@ def test_complex_synonyms(self):
             model.save_pretrained(tmpdir_in)
 
             main_export(model_name_or_path=tmpdir_in, output=tmpdir_out, task="text-classification")
+
+    @parameterized.expand(_get_models_to_test(PYTORCH_REMOTE_CODE_MODELS, library_name="transformers"))
+    @require_torch
+    @slow
+    @pytest.mark.run_slow
+    def test_custom_model(
+            self,
+            test_name: str,
+            model_type: str,
+            model_name: str,
+            task: str,
+            variant: str,
+            monolith: bool,
+            no_post_process: bool,
+    ):
+        with TemporaryDirectory() as tmpdirname:
+            out = subprocess.run(
+                f"python3 -m optimum.exporters.onnx --trust-remote-code --model {model_name} --task {task} {tmpdirname}",
+                shell=True,
+                capture_output=True,
+            )
+            self.assertFalse(out.returncode)
+
+        with TemporaryDirectory() as tmpdirname:
+            out = subprocess.run(
+                f"python3 -m optimum.exporters.onnx --trust-remote-code --model {model_name} --task {task} {tmpdirname}",
+                shell=True,
+                check=True,
+            )

tests/exporters/onnx/test_onnx_export.py

@@ -51,6 +51,7 @@
     PYTORCH_EXPORT_MODELS_TINY,
     PYTORCH_SENTENCE_TRANSFORMERS_MODEL,
     PYTORCH_STABLE_DIFFUSION_MODEL,
+    PYTORCH_REMOTE_CODE_MODELS,
     PYTORCH_TIMM_MODEL,
     TENSORFLOW_EXPORT_MODELS,
     VALIDATE_EXPORT_ON_SHAPES_SLOW,
@@ -197,7 +198,8 @@ def _onnx_export(
             return
 
         onnx_config = onnx_config_class_constructor(model.config)
-
+        print("Got this ONNX config")
+        print(onnx_config)
         # We need to set this to some value to be able to test the outputs values for batch size > 1.
         if (
             isinstance(onnx_config, OnnxConfigWithPast)
@@ -323,6 +325,7 @@ def test_all_models_tested(self):
             - set(PYTORCH_EXPORT_MODELS_TINY.keys())
             - set(PYTORCH_TIMM_MODEL.keys())
             - set(PYTORCH_SENTENCE_TRANSFORMERS_MODEL.keys())
+            - set(PYTORCH_REMOTE_CODE_MODELS.keys())
         )
         if len(missing_models_set) > 0:
             self.fail(f"Not testing all models. Missing models: {missing_models_set}")

tests/onnxruntime/test_modeling.py

@@ -1987,6 +1987,7 @@ class ORTModelForFeatureExtractionIntegrationTest(ORTModelTestMixin):
         "mpnet",
         "roberta",
         "xlm_roberta",
+        "nomic-bert"
     ]
 
     FULL_GRID = {"model_arch": SUPPORTED_ARCHITECTURES}
@@ -4703,7 +4704,7 @@ def test_generate_utils(self, test_name: str, model_arch: str, use_cache: str):
     @parameterized.expand(grid_parameters(FULL_GRID))
     def test_compare_to_transformers(self, test_name: str, model_arch: str, use_cache: bool, use_merged: bool):
         if use_cache is False and use_merged is True:
-            self.skipTest("use_cache=False, use_merged=True are uncompatible")
+            self.skipTest("use_cache=False, use_merged=True are incompatible")
 
         model_args = {
             "test_name": test_name,