Switches to 4.56.1 and other changes (#216)

* switch to 4.56.1 and other changes

* fix

Author: xadupre

.github/workflows/ci.yml

@@ -16,7 +16,7 @@ jobs:
       matrix:
         os: [ubuntu-latest]
         python: ['3.10', '3.11', '3.12', '3.13']
-        transformers: ['4.48.3', '4.51.3', '4.52.4', '4.53.3', '4.55.4', '4.56.0', 'main']
+        transformers: ['4.48.3', '4.51.3', '4.52.4', '4.53.3', '4.55.4', '4.56.1', 'main']
         torch: ['2.8', 'main']
         exclude:
           - python: '3.10'
@@ -30,7 +30,7 @@ jobs:
           - python: '3.10'
             transformers: '4.55.4'
           - python: '3.10'
-            transformers: '4.56.0'
+            transformers: '4.56.1'
           - python: '3.11'
             torch: 'main'
           - python: '3.11'
@@ -40,7 +40,7 @@ jobs:
           - python: '3.11'
             transformers: '4.55.4'
           - python: '3.11'
-            transformers: '4.56.0'
+            transformers: '4.56.1'
           - python: '3.13'
             torch: '2.8'
           - python: '3.13'

_unittests/ut_tasks/try_tasks.py

@@ -755,6 +755,42 @@ def test_imagetext2text_generation_zai_glm(self):
         )
         print(output_text)
 
+    @never_test()
+    def test_sentence_similary_alibaba_nlp_gte(self):
+        """
+        clear&&NEVERTEST=1 python _unittests/ut_tasks/try_tasks.py -k alibaba
+        """
+        import torch.nn.functional as F
+        from transformers import AutoModel, AutoTokenizer
+
+        input_texts = [
+            "what is the capital of China?",
+            "how to implement quick sort in python?",
+            "Beijing",
+            "sorting algorithms",
+        ]
+
+        model_path = "Alibaba-NLP/gte-large-en-v1.5"
+        tokenizer = AutoTokenizer.from_pretrained(model_path)
+        model = AutoModel.from_pretrained(model_path, trust_remote_code=True)
+
+        # Tokenize the input texts
+        batch_dict = tokenizer(
+            input_texts, max_length=8192, padding=True, truncation=True, return_tensors="pt"
+        )
+
+        print("-- type:", type(model))
+        print("-- subclasses:", type(model).__subclasses__())
+        print("-- inputs:", self.string_type(batch_dict, with_shape=True))
+        outputs = model(**batch_dict)
+        print("-- outputs:", self.string_type(outputs, with_shape=True))
+        embeddings = outputs.last_hidden_state[:, 0]
+
+        # (Optionally) normalize embeddings
+        embeddings = F.normalize(embeddings, p=2, dim=1)
+        scores = (embeddings[:1] @ embeddings[1:].T) * 100
+        print(scores.tolist())
+
 
 if __name__ == "__main__":
     unittest.main(verbosity=2)

onnx_diagnostic/helpers/helper.py

@@ -774,6 +774,14 @@ def string_type(
             return f"{obj.__class__.__name__}(**{s})"
     if obj.__class__.__name__ in {"TorchModelContainer", "InferenceSession"}:
         return f"{obj.__class__.__name__}(...)"
+    if obj.__class__.__name__ == "Results":
+        import ultralytics
+
+        assert isinstance(
+            obj, ultralytics.engine.results.Results
+        ), f"Unexpected type={type(obj)}"
+        return f"ultralytics.{obj.__class__.__name__}(...)"
+
     if verbose:
         print(f"[string_type] END:{type(obj)}")
     raise AssertionError(f"Unsupported type {type(obj).__name__!r} - {type(obj)}")

onnx_diagnostic/helpers/torch_helper.py

@@ -721,9 +721,10 @@ def to_any(value: Any, to_value: Union[torch.dtype, torch.device, str]) -> Any:
         return {to_any(t, to_value) for t in value}
     if type(value) is dict:
         return {k: to_any(t, to_value) for k, t in value.items()}
-    if value.__class__.__name__ == "DynamicCache":
+    if value.__class__.__name__ in {"DynamicCache", "HybridCache"}:
+        make = dict(DynamicCache=make_dynamic_cache, HybridCache=make_hybrid_cache)
         cc = CacheKeyValue(value)
-        return make_dynamic_cache(
+        return make[value.__class__.__name__](  # type: ignore[operator]
             list(
                 zip(
                     [t.to(to_value) if t is not None else t for t in cc.key_cache],
@@ -822,6 +823,15 @@ def torch_deepcopy(value: Any) -> Any:
         new_args = torch_deepcopy(args)
         return torch.utils._pytree.tree_unflatten(new_args, spec)
 
+    if value.__class__.__name__ == "Results":
+        import copy
+        import ultralytics
+
+        assert isinstance(
+            value, ultralytics.engine.results.Results
+        ), f"Unexpected type={type(value)}"
+        return copy.deepcopy(value)
+
     # We should have a code using serialization, deserialization assuming a model
     # cannot be exported without them.
     raise NotImplementedError(f"torch_deepcopy not implemented for type {type(value)}")

onnx_diagnostic/torch_models/hghub/hub_data.py

@@ -11,6 +11,7 @@
     """
     architecture,task
     ASTModel,feature-extraction
+    AutoencoderKL,image-to-image
     AlbertModel,feature-extraction
     BeitForImageClassification,image-classification
     BartForConditionalGeneration,summarization
@@ -154,6 +155,7 @@
     Wav2Vec2ForCTC,automatic-speech-recognition
     YolosForObjectDetection,object-detection
     YolosModel,image-feature-extraction
+    Alibaba-NLP/gte-large-en-v1.5,sentence-similarity
     emilyalsentzer/Bio_ClinicalBERT,fill-mask"""
 )