fix type hint error, vison language models tests and phi3

Author: IlyasMoutawwakil

optimum/exporters/openvino/model_patcher.py

@@ -1581,19 +1581,19 @@ def __enter__(self):
         ):
             self._model.config.max_position_embeddings = self._model.config.original_max_position_embeddings
 
-        if is_transformers_version(">=", "4.42.0"):
+        if is_transformers_version(">=", "4.42.0") and is_transformers_version("<", "4.48.0"):
             self._model.model._orig_forward = self._model.model.forward
             self._model.model.forward = types.MethodType(phi3_442_forward, self._model.model)
 
         # https://github.com/huggingface/transformers/blob/30ee508c6c92a1c0aa0281d193c7c0fb815b8d2f/src/transformers/models/phi3/modeling_phi3.py#L113
         # init inv_freq for torchscript tracing
         for layer in self._model.model.layers:
-            if is_torch_version(">=", "2.1.0"):
+            if is_torch_version(">=", "2.1.0") and is_transformers_version("<", "4.48.0"):
                 orig_self_attn_fwd = layer.self_attn.forward
                 layer.self_attn.forward = types.MethodType(_phi3_self_attn_sdpa_forward, layer.self_attn)
                 layer.self_attn._orig_forward = orig_self_attn_fwd
 
-            if layer.self_attn.rotary_emb.inv_freq is None:
+            if hasattr(layer.self_attn, "rotary_emb") and layer.self_attn.rotary_emb.inv_freq is None:
                 rotary_emb = layer.self_attn.rotary_emb
                 layer.self_attn.rotary_emb.inv_freq = 1.0 / (
                     rotary_emb.base ** (torch.arange(0, rotary_emb.dim, 2, dtype=torch.int64).float() / rotary_emb.dim)

optimum/intel/openvino/modeling_decoder.py

@@ -56,7 +56,6 @@
 
 
 if TYPE_CHECKING:
-    from transformers.generation.streamers import BaseStreamer
     from transformers.modeling_utils import PreTrainedModel
 
 
@@ -706,7 +705,6 @@ def generate(
         prefix_allowed_tokens_fn: Optional[Callable[[int, torch.Tensor], List[int]]] = None,
         synced_gpus: Optional[bool] = None,
         assistant_model: Optional["PreTrainedModel"] = None,
-        streamer: Optional["BaseStreamer"] = None,
         negative_prompt_ids: Optional[torch.Tensor] = None,
         negative_prompt_attention_mask: Optional[torch.Tensor] = None,
         **kwargs,
@@ -734,7 +732,6 @@ def generate(
             prefix_allowed_tokens_fn,
             synced_gpus,
             assistant_model,
-            streamer,
             negative_prompt_ids,
             negative_prompt_attention_mask,
             **kwargs,

tests/openvino/test_modeling.py

@@ -2172,6 +2172,11 @@ def test_compare_to_transformers(self, model_arch):
         for component_name, component in ov_model.components.items():
             self.assertIsInstance(component, MODEL_PARTS_CLS_MAPPING[component_name])
         self.assertIsInstance(ov_model.config, PretrainedConfig)
+
+        # TODO: fix in models
+        if preprocessors.get("processor") is not None:
+            preprocessors["processor"].patch_size = ov_model.config.vision_config.patch_size
+
         inputs = ov_model.preprocess_inputs(**preprocessors, text=prompt, image=self.IMAGE.resize((600, 600)))
         transformers_inputs = copy.deepcopy(inputs)
         test_device = "AUTO"