add orion

Author: eaidova

optimum/exporters/openvino/__main__.py

@@ -203,6 +203,8 @@ def main_export(
         do_gptq_patching = quantization_config and quantization_config["quant_method"] == "gptq"
         model_type = config.model_type.replace("_", "-")
 
+        if model_type in {"falcon", "mpt"} and trust_remote_code:
+            trust_remote_code = False
         if model_type not in TasksManager._SUPPORTED_MODEL_TYPE:
             custom_architecture = True
         elif task not in TasksManager.get_supported_tasks_for_model_type(

optimum/exporters/openvino/model_configs.py

@@ -75,7 +75,7 @@ def init_model_configs():
 
 
 @register_in_tasks_manager("baichuan", *["text-generation", "text-generation-with-past"], library_name="transformers")
-class BaichaunOpenVINOConfig(TextDecoderOnnxConfig):
+class BaichaunOpenVINOConfig(TextDecoderWithPositionIdsOnnxConfig):
     DEFAULT_ONNX_OPSET = 13
     NORMALIZED_CONFIG_CLASS = NormalizedTextConfig.with_args(
         num_layers="num_hidden_layers", num_attention_heads="num_attention_heads", hidden_size="hidden_size"
@@ -471,3 +471,12 @@ class DeciOpenVINOConfig(TextDecoderWithPositionIdsOnnxConfig):
     DUMMY_INPUT_GENERATOR_CLASSES = (DummyTextInputGenerator, DeciDummyPastKeyValuesGenerator)
     DUMMY_PKV_GENERATOR_CLASS = DeciDummyPastKeyValuesGenerator
     NORMALIZED_CONFIG_CLASS = NormalizedTextConfig
+
+
+@register_in_tasks_manager("orion", *["text-generation", "text-generation-with-past"], library_name="transformers")
+class OrionOpenVINOConfig(TextDecoderWithPositionIdsOnnxConfig):
+    DEFAULT_ONNX_OPSET = 14
+
+    DUMMY_INPUT_GENERATOR_CLASSES = (DummyTextInputGenerator, MistralDummyPastKeyValuesGenerator)
+    DUMMY_PKV_GENERATOR_CLASS = MistralDummyPastKeyValuesGenerator
+    NORMALIZED_CONFIG_CLASS = NormalizedTextConfig

tests/openvino/test_modeling.py

@@ -524,10 +524,11 @@ class OVModelForCausalLMIntegrationTest(unittest.TestCase):
         "stablelm",
         "starcoder2",
         "phi",
+        "internlm2",
     )
     GENERATION_LENGTH = 100
     IS_SUPPORT_STATEFUL = is_openvino_version(">=", "2023.3")
-    REMOTE_CODE_MODELS = ("chatglm", "minicpm", "baichuan2", "jais", "qwen")
+    REMOTE_CODE_MODELS = ("chatglm", "minicpm", "baichuan2", "jais", "qwen", "internlm2", "decilm")
 
     @parameterized.expand(SUPPORTED_ARCHITECTURES)
     def test_compare_to_transformers(self, model_arch):

tests/openvino/utils_tests.py

@@ -36,6 +36,7 @@
     "data2vec_audio": "hf-internal-testing/tiny-random-Data2VecAudioModel",
     "deberta": "hf-internal-testing/tiny-random-deberta",
     "deberta_v2": "hf-internal-testing/tiny-random-DebertaV2Model",
+    "decilm": "katuni4ka/tiny-random-deciml",
     "deit": "hf-internal-testing/tiny-random-deit",
     "convnext": "hf-internal-testing/tiny-random-convnext",
     "distilbert": "hf-internal-testing/tiny-random-distilbert",
@@ -50,6 +51,7 @@
     "gptj": "hf-internal-testing/tiny-random-GPTJModel",
     "hubert": "hf-internal-testing/tiny-random-HubertModel",
     "ibert": "hf-internal-testing/tiny-random-ibert",
+    "internlm2": "katuni4ka/tiny-random-internlm2",
     "levit": "hf-internal-testing/tiny-random-LevitModel",
     "longt5": "hf-internal-testing/tiny-random-longt5",
     "llama": "fxmarty/tiny-llama-fast-tokenizer",