Cover more models with openvino export (#709)

* cover more models with openvino export

* xglm

* fix tests

Author: eaidova

optimum/exporters/openvino/model_configs.py

@@ -592,3 +592,58 @@ def outputs(self) -> Dict[str, Dict[int, str]]:
         return {
             "sample": {0: "batch_size", 2: "height", 3: "width"},
         }
+
+
+@register_in_tasks_manager(
+    "persimmon",
+    *[
+        "feature-extraction",
+        "feature-extraction-with-past",
+        "text-generation",
+        "text-generation-with-past",
+        "text-classification",
+    ],
+    library_name="transformers",
+)
+class PersimmonOpenVINOConfig(TextDecoderWithPositionIdsOnnxConfig):
+    DEFAULT_ONNX_OPSET = 14
+    NORMALIZED_CONFIG_CLASS = NormalizedTextConfig
+
+
+@register_in_tasks_manager("biogpt", *["text-generation", "text-generation-with-past"], library_name="transformers")
+class BioGPTOpenVINOConfig(TextDecoderOnnxConfig):
+    # BioGPT does not require position_ids input.
+    DEFAULT_ONNX_OPSET = 13
+    NORMALIZED_CONFIG_CLASS = NormalizedTextConfig
+
+
+@register_in_tasks_manager(
+    "gpt-neox-japanese", *["text-generation", "text-generation-with-past"], library_name="transformers"
+)
+class GPTNeoxJapaneseOpenVINOConfig(TextDecoderOnnxConfig):
+    # GPTNeoxJapanese does not require position_ids input.
+    DEFAULT_ONNX_OPSET = 13
+    NORMALIZED_CONFIG_CLASS = NormalizedTextConfig
+
+
+@register_in_tasks_manager(
+    "cohere",
+    *[
+        "feature-extraction",
+        "feature-extraction-with-past",
+        "text-generation",
+        "text-generation-with-past",
+        "text-classification",
+    ],
+    library_name="transformers",
+)
+class CohereOpenVINOConfig(LlamaOpenVINOConfig):
+    pass
+
+
+@register_in_tasks_manager("xglm", *["text-generation", "text-generation-with-past"], library_name="transformers")
+class XGLMConfig(TextDecoderWithPositionIdsOnnxConfig):
+    DEFAULT_ONNX_OPSET = 13
+    NORMALIZED_CONFIG_CLASS = NormalizedTextConfig.with_args(
+        num_attention_heads="attention_heads", hidden_size="d_model"
+    )

tests/openvino/test_modeling.py

@@ -552,6 +552,11 @@ class OVModelForCausalLMIntegrationTest(unittest.TestCase):
         "orion",
         "falcon",
         "falcon-40b",
+        "persimmon",
+        "biogpt",
+        "gpt_neox_japanese",
+        "cohere",
+        "xglm",
     )
     GENERATION_LENGTH = 100
     REMOTE_CODE_MODELS = (
@@ -617,8 +622,11 @@ def test_compare_to_transformers(self, model_arch):
         if model_arch == "qwen":
             return
 
-        if model_arch != "chatglm":
+        if model_arch not in ["chatglm", "persimmon"]:
             tokenizer.pad_token_id = tokenizer.eos_token_id
+
+        if model_arch == "persimmon":
+            tokenizer.pad_token_id = tokenizer.bos_token_id
         # Compare batched generation
         tokenizer.padding_side = "left"
         tokens = tokenizer(["Today is a nice day and I am longer", "This is me"], return_tensors="pt", padding=True)

tests/openvino/utils_tests.py

@@ -26,11 +26,13 @@
     "baichuan2": "katuni4ka/tiny-random-baichuan2",
     "baichuan2-13b": "katuni4ka/tiny-random-baichuan2-13b",
     "bigbird_pegasus": "hf-internal-testing/tiny-random-bigbird_pegasus",
+    "biogpt": "hf-tiny-model-private/tiny-random-BioGptForCausalLM",
     "blenderbot-small": "hf-internal-testing/tiny-random-BlenderbotModel",
     "blenderbot": "hf-internal-testing/tiny-random-BlenderbotModel",
     "bloom": "hf-internal-testing/tiny-random-BloomModel",
     "camembert": "hf-internal-testing/tiny-random-camembert",
     "convbert": "hf-internal-testing/tiny-random-ConvBertForSequenceClassification",
+    "cohere": "hf-internal-testing/tiny-random-CohereForCausalLM",
     "chatglm": "katuni4ka/tiny-random-chatglm2",
     "codegen": "hf-internal-testing/tiny-random-CodeGenForCausalLM",
     "data2vec_text": "hf-internal-testing/tiny-random-Data2VecTextModel",
@@ -51,6 +53,7 @@
     "gpt2": "hf-internal-testing/tiny-random-gpt2",
     "gpt_neo": "hf-internal-testing/tiny-random-GPTNeoModel",
     "gpt_neox": "hf-internal-testing/tiny-random-GPTNeoXForCausalLM",
+    "gpt_neox_japanese": "hf-internal-testing/tiny-random-GPTNeoXJapaneseForCausalLM",
     "gptj": "hf-internal-testing/tiny-random-GPTJModel",
     "hubert": "hf-internal-testing/tiny-random-HubertModel",
     "ibert": "hf-internal-testing/tiny-random-ibert",
@@ -78,6 +81,7 @@
     "olmo": "katuni4ka/tiny-random-olmo-hf",
     "orion": "katuni4ka/tiny-random-orion",
     "pegasus": "hf-internal-testing/tiny-random-pegasus",
+    "persimmon": "hf-internal-testing/tiny-random-PersimmonForCausalLM",
     "pix2struct": "fxmarty/pix2struct-tiny-random",
     "phi": "echarlaix/tiny-random-PhiForCausalLM",
     "phi3": "katuni4ka/tiny-random-phi3",
@@ -115,6 +119,7 @@
     "whisper": "openai/whisper-tiny.en",
     "xlm": "hf-internal-testing/tiny-random-xlm",
     "xlm_roberta": "hf-internal-testing/tiny-xlm-roberta",
+    "xglm": "hf-internal-testing/tiny-random-XGLMForCausalLM",
 }