add hub_kwargs and default pipeline tests

Author: jiqing-feng

optimum/exporters/openvino/model_patcher.py

@@ -327,9 +327,9 @@ def _llama_gemma_update_causal_mask(self, attention_mask, input_tensor, cache_po
                 offset = 0
             mask_shape = attention_mask.shape
             mask_slice = (attention_mask.eq(0.0)).to(dtype=dtype) * min_dtype
-            causal_mask[
-                : mask_shape[0], : mask_shape[1], offset : mask_shape[2] + offset, : mask_shape[3]
-            ] = mask_slice
+            causal_mask[: mask_shape[0], : mask_shape[1], offset : mask_shape[2] + offset, : mask_shape[3]] = (
+                mask_slice
+            )
 
     if (
         self.config._attn_implementation == "sdpa"

optimum/intel/pipelines/pipeline_base.py

@@ -99,6 +99,7 @@ def load_ipex_model(
     targeted_task,
     SUPPORTED_TASKS,
     model_kwargs: Optional[Dict[str, Any]] = None,
+    hub_kwargs: Optional[Dict[str, Any]] = None,
     **kwargs,
 ):
     export = kwargs.pop("export", True)
@@ -109,7 +110,7 @@ def load_ipex_model(
 
     if model is None:
         model_id = SUPPORTED_TASKS[targeted_task]["default"]
-        model = ipex_model_class.from_pretrained(model_id, export=True, **model_kwargs)
+        model = ipex_model_class.from_pretrained(model_id, export=True, **model_kwargs, **hub_kwargs)
     elif isinstance(model, str):
         model_id = model
         try:
@@ -121,7 +122,7 @@ def load_ipex_model(
                 "config file not found, please pass `export` to decide whether we should export this model. `export` defaullt to True"
             )
 
-        model = ipex_model_class.from_pretrained(model, export=export, **model_kwargs)
+        model = ipex_model_class.from_pretrained(model, export=export, **model_kwargs, **hub_kwargs)
     elif isinstance(model, IPEXModel):
         model_id = None
     else:
@@ -280,7 +281,9 @@ def pipeline(
 
     # Load the correct model if possible
     # Infer the framework from the model if not already defined
-    model, model_id = MAPPING_LOADING_FUNC[accelerator](model, task, supported_tasks, model_kwargs, **kwargs)
+    model, model_id = MAPPING_LOADING_FUNC[accelerator](
+        model, task, supported_tasks, model_kwargs, hub_kwargs, **kwargs
+    )
 
     if load_tokenizer and model_id and tokenizer is None:
         tokenizer = AutoTokenizer.from_pretrained(model_id, **hub_kwargs, **model_kwargs)

tests/ipex/test_pipelines.py

@@ -22,6 +22,7 @@
 from transformers.pipelines import pipeline as transformers_pipeline
 
 from optimum.intel.ipex.modeling_base import (
+    IPEXModel,
     IPEXModelForAudioClassification,
     IPEXModelForCausalLM,
     IPEXModelForImageClassification,
@@ -122,6 +123,15 @@ class PipelinesIntegrationTest(unittest.TestCase):
         "resnet",
         "vit",
     )
+    SUPPORT_TASKS = (
+        "text-generation",
+        "fill-mask",
+        "question-answering",
+        "image-classification",
+        "text-classification",
+        "token-classification",
+        "audio-classification",
+    )
 
     @parameterized.expand(COMMON_SUPPORTED_ARCHITECTURES)
     def test_token_classification_pipeline_inference(self, model_arch):
@@ -263,3 +273,8 @@ def test_pipeline_load_from_jit_model(self, model_arch):
         self.assertTrue(isinstance(ipex_generator.model, IPEXModelForSequenceClassification))
         self.assertTrue(isinstance(ipex_generator.model.model, torch.jit.RecursiveScriptModule))
         self.assertGreaterEqual(ipex_output[0]["score"], 0.0)
+
+    @parameterized.expand(SUPPORT_TASKS)
+    def test_pipeline_with_default_model(self, task):
+        ipex_generator = ipex_pipeline(task, accelerator="ipex")
+        self.assertTrue(isinstance(ipex_generator.model, IPEXModel))