onnx-diagnostic 0.7.6__py3-none-any.whl → 0.7.7__py3-none-any.whl

onnx_diagnostic/tasks/mixture_of_expert.py

@@ -2,7 +2,7 @@ from typing import Any, Callable, Dict, Optional, Tuple
 import torch
 
 # from ..helpers.cache_helper import make_dynamic_cache
-from ..helpers.config_helper import update_config  # , check_hasattr, _pick
+from ..helpers.config_helper import update_config, default_num_hidden_layers as nhl
 
 __TASK__ = "MoE"
 
@@ -11,7 +11,7 @@ def reduce_model_config(config: Any) -> Dict[str, Any]:
     """Reduces a model size."""
     kwargs: Dict[str, Any] = {}
     if hasattr(config, "num_hidden_layers"):
-        config.num_hidden_layers = min(config.num_hidden_layers, 2)
+        config.num_hidden_layers = min(config.num_hidden_layers, nhl())
     if hasattr(config, "vision_config") and hasattr(config.vision_config, "num_hidden_layers"):
         config.vision_config.num_hidden_layers = min(config.vision_config.num_hidden_layers, 2)
     if hasattr(config, "audio_processor") and hasattr(

onnx_diagnostic/tasks/object_detection.py

@@ -1,6 +1,10 @@
 from typing import Any, Callable, Dict, Optional, Tuple
 import torch
-from ..helpers.config_helper import update_config, check_hasattr
+from ..helpers.config_helper import (
+    update_config,
+    check_hasattr,
+    default_num_hidden_layers as nhl,
+)
 
 __TASK__ = "object-detection"
 
@@ -10,7 +14,7 @@ def reduce_model_config(config: Any) -> Dict[str, Any]:
     check_hasattr(config, ("num_hidden_layers", "hidden_sizes"))
     kwargs = dict(
         num_hidden_layers=(
-            min(config.num_hidden_layers, 2)
+            min(config.num_hidden_layers, nhl())
             if hasattr(config, "num_hidden_layers")
             else len(config.hidden_sizes)
         )

onnx_diagnostic/tasks/sentence_similarity.py

@@ -1,6 +1,10 @@
 from typing import Any, Callable, Dict, Optional, Tuple
 import torch
-from ..helpers.config_helper import update_config, check_hasattr
+from ..helpers.config_helper import (
+    update_config,
+    check_hasattr,
+    default_num_hidden_layers as nhl,
+)
 
 __TASK__ = "sentence-similarity"
 
@@ -9,7 +13,7 @@ def reduce_model_config(config: Any) -> Dict[str, Any]:
     """Reduces a model size."""
     check_hasattr(config, "num_attention_heads", "num_hidden_layers")
     kwargs = dict(
-        num_hidden_layers=min(config.num_hidden_layers, 2),
+        num_hidden_layers=min(config.num_hidden_layers, nhl()),
         num_attention_heads=min(config.num_attention_heads, 4),
     )
     update_config(config, kwargs)

onnx_diagnostic/tasks/summarization.py

@@ -1,7 +1,12 @@
 from typing import Any, Callable, Dict, Optional, Tuple
 import torch
 from ..helpers.cache_helper import make_dynamic_cache, make_encoder_decoder_cache
-from ..helpers.config_helper import update_config, check_hasattr, _pick
+from ..helpers.config_helper import (
+    update_config,
+    check_hasattr,
+    _pick,
+    default_num_hidden_layers as nhl,
+)
 
 __TASK__ = "summarization"
 
@@ -12,7 +17,7 @@ def reduce_model_config(config: Any) -> Dict[str, Any]:
     if hasattr(config, "num_decoder_layers"):
         config.num_decoder_layers = min(config.num_decoder_layers, 2)
     if hasattr(config, "num_hidden_layers"):
-        config.num_hidden_layers = min(config.num_hidden_layers, 2)
+        config.num_hidden_layers = min(config.num_hidden_layers, nhl())
     update_config(config, kwargs)
     return kwargs
 

onnx_diagnostic/tasks/text2text_generation.py

@@ -1,7 +1,12 @@
 from typing import Any, Callable, Dict, Optional, Tuple
 import torch
 from ..helpers.cache_helper import make_dynamic_cache, make_encoder_decoder_cache
-from ..helpers.config_helper import update_config, check_hasattr, _pick
+from ..helpers.config_helper import (
+    update_config,
+    check_hasattr,
+    _pick,
+    default_num_hidden_layers as nhl,
+)
 
 __TASK__ = "text2text-generation"
 
@@ -12,7 +17,7 @@ def reduce_model_config(config: Any) -> Dict[str, Any]:
     if hasattr(config, "num_decoder_layers"):
         config.num_decoder_layers = min(config.num_decoder_layers, 2)
     if hasattr(config, "num_hidden_layers"):
-        config.num_hidden_layers = min(config.num_hidden_layers, 2)
+        config.num_hidden_layers = min(config.num_hidden_layers, nhl())
     update_config(config, kwargs)
     return kwargs
 

onnx_diagnostic/tasks/text_classification.py

@@ -1,6 +1,10 @@
 from typing import Any, Callable, Dict, Optional, Tuple
 import torch
-from ..helpers.config_helper import update_config, check_hasattr
+from ..helpers.config_helper import (
+    update_config,
+    check_hasattr,
+    default_num_hidden_layers as nhl,
+)
 
 __TASK__ = "text-classification"
 
@@ -9,7 +13,7 @@ def reduce_model_config(config: Any) -> Dict[str, Any]:
     """Reduces a model size."""
     check_hasattr(config, "num_attention_heads", "num_hidden_layers")
     kwargs = dict(
-        num_hidden_layers=min(config.num_hidden_layers, 2),
+        num_hidden_layers=min(config.num_hidden_layers, nhl()),
         num_attention_heads=min(config.num_attention_heads, 4),
     )
     update_config(config, kwargs)

onnx_diagnostic/tasks/text_generation.py

@@ -6,7 +6,12 @@ from ..helpers.cache_helper import (
     make_sliding_window_cache,
     make_static_cache,
 )
-from ..helpers.config_helper import update_config, check_hasattr, _pick
+from ..helpers.config_helper import (
+    update_config,
+    check_hasattr,
+    _pick,
+    default_num_hidden_layers as nhl,
+)
 
 __TASK__ = "text-generation"
 
@@ -25,7 +30,7 @@ def reduce_model_config(config: Any) -> Dict[str, Any]:
     if config.__class__.__name__ == "FalconMambaConfig":
         check_hasattr(config, "conv_kernel", "state_size", "intermediate_size")  # 4 and 8
         kwargs = dict(
-            num_hidden_layers=min(config.num_hidden_layers, 2),
+            num_hidden_layers=min(config.num_hidden_layers, nhl()),
             intermediate_size=256 if config is None else min(512, config.intermediate_size),
             hidden_size=512 if config is None else min(512, config.hidden_size),
             cls_cache="MambaCache",
@@ -37,24 +42,13 @@ def reduce_model_config(config: Any) -> Dict[str, Any]:
             head_dim=getattr(
                 config, "head_dim", config.hidden_size // config.num_attention_heads
             ),
-            num_hidden_layers=min(config.num_hidden_layers, 2),
+            num_hidden_layers=min(config.num_hidden_layers, nhl()),
             num_key_value_heads=(
                 config.num_key_value_heads
                 if hasattr(config, "num_key_value_heads")
                 else config.num_attention_heads
             ),
-            hidden_size=(
-                min(config.hidden_size, 4096 // 4)
-                if config.hidden_size % 64 == 0
-                else config.hidden_size
-            ),
         )
-        if config is None or hasattr(config, "intermediate_size"):
-            kwargs["intermediate_size"] = (
-                min(config.intermediate_size, 24576 // 4)
-                if config.intermediate_size % 4 == 0
-                else config.intermediate_size
-            )
     update_config(config, kwargs)
     return kwargs
 

onnx_diagnostic/torch_export_patches/onnx_export_errors.py

@@ -361,7 +361,7 @@ def torch_export_patches(
             torch._meta_registrations._broadcast_shapes = patched__broadcast_shapes
 
         # torch._export.non_strict_utils.produce_guards_and_solve_constraints
-        if catch_constraints:
+        if patch_torch and catch_constraints:
             if verbose:
                 print("[torch_export_patches] modifies shape constraints")
             f_produce_guards_and_solve_constraints = (
@@ -513,7 +513,7 @@ def torch_export_patches(
                 if verbose:
                     print("[torch_export_patches] restored pytorch functions")
 
-            if stop_if_static:
+            if patch_torch and stop_if_static:
                 if verbose:
                     print("[torch_export_patches] restored ShapeEnv._set_replacement")
 
@@ -529,7 +529,7 @@ def torch_export_patches(
                         print("[torch_export_patches] restored ShapeEnv._check_frozen")
                     ShapeEnv._check_frozen = f_shape_env__check_frozen
 
-            if catch_constraints:
+            if patch_torch and catch_constraints:
                 # to catch or skip dynamic_shapes issues
                 torch._export.non_strict_utils.produce_guards_and_solve_constraints = (
                     f_produce_guards_and_solve_constraints

onnx_diagnostic/torch_export_patches/patch_inputs.py

@@ -34,7 +34,7 @@ def _make_shape(subset: Dict, cls: type, value: Any) -> Any:
             f"Inconsistencies in subset={subset}, found={values}, "
             f"it cannot be a {cls}, value={string_type(value)}"
         )
-        cache_length = len(value.key_cache)
+        cache_length = len(value.layers if hasattr(value, "layers") else value.key_cache)
         for v in subset.values():
             axes = v
             break

onnx_diagnostic/torch_export_patches/patches/patch_torch.py

@@ -27,8 +27,8 @@ def _catch_produce_guards_and_solve_constraints(
     dynamic_shapes: Union[Dict[str, Any], Tuple[Any], List[Any], None],
     equalities_inputs: "EqualityConstraint",  # noqa: F821
     original_signature: inspect.Signature,
-    _is_torch_jit_trace: bool = False,
     verbose: int = 0,
+    **kwargs,
 ):
     try:
         return previous_function(
@@ -37,7 +37,7 @@ def _catch_produce_guards_and_solve_constraints(
             dynamic_shapes=dynamic_shapes,
             equalities_inputs=equalities_inputs,
             original_signature=original_signature,
-            _is_torch_jit_trace=_is_torch_jit_trace,
+            **kwargs,
         )
     except Exception as e:
         if not int(os.environ.get("SKIP_SOLVE_CONSTRAINTS", "1")):
@@ -51,7 +51,7 @@ def _catch_produce_guards_and_solve_constraints(
                 f"dynamic_shapes={dynamic_shapes}\n"
                 f"equalities_inputs={equalities_inputs}\n"
                 f"original_signature={original_signature}\n"
-                f"_is_torch_jit_trace={_is_torch_jit_trace}\n"
+                f"kwargs={kwargs}\n"
                 f"exc={e}\ngm={gm}"
             )
         torch._dynamo.reset()
@@ -174,7 +174,7 @@ class patched_ShapeEnv:
             self.counter["ignored_backward_guard"] += 1
             raise AssertionError(
                 f"[patched_ShapeEnv] Ignored guard {expr} == {concrete_val}, "
-                f"this could result in accuracy problems."
+                f"this could result in accuracy problems"
             )
 
     def _set_replacement(

onnx_diagnostic/torch_export_patches/patches/patch_transformers.py

@@ -1,4 +1,5 @@
 import inspect
+import math
 from dataclasses import dataclass
 from functools import wraps
 from typing import Callable, List, Optional, Tuple
@@ -23,6 +24,14 @@ except ImportError:
     patch_masking_utils = False
 
 
+try:
+    # transformers>= 4.55.1
+    from transformers.cache_utils import DynamicLayer
+
+    patch_DynamicLayer = hasattr(DynamicLayer, "lazy_initialization")
+except ImportError:
+    patch_DynamicLayer = False
+
 from ...ext_test_case import has_transformers
 from ...helpers.torch_helper import is_torchdynamo_exporting
 
@@ -157,6 +166,20 @@ if patch_parse_processor_args:
         return processor_kwargs, remaining_kwargs
 
 
+if patch_DynamicLayer:
+
+    class patched_DynamicLayer:
+        _PATCHES_ = ["lazy_initialization"]
+        _PATCHED_CLASS_ = DynamicLayer
+
+        def lazy_initialization(self, key_states: torch.Tensor):
+            self.dtype, self.device = key_states.dtype, key_states.device
+            new_shape = list(key_states.shape)
+            new_shape[-2] = 0
+            self.keys = torch.empty(new_shape, dtype=self.dtype, device=self.device)
+            self.values = torch.empty(new_shape, dtype=self.dtype, device=self.device)
+
+
 def _patch_make_causal_mask(
     input_ids_shape: torch.Size,
     dtype: torch.dtype,
@@ -323,6 +346,14 @@ if pv.Version(transformers.__version__) < pv.Version("4.51"):
                     self.key_cache[layer_idx] = key_states
                     self.value_cache[layer_idx] = value_states
                 else:
+                    torch._check(
+                        len(self.key_cache[layer_idx].shape) == len(key_states.shape),
+                        lambda: (
+                            f"Rank mismatch len(self.key_cache[layer_idx].shape)="
+                            f"{len(self.key_cache[layer_idx].shape)}, "
+                            f"len(key_states.shape)={len(key_states.shape)}"
+                        ),
+                    )
                     self.key_cache[layer_idx] = torch.cat(
                         [self.key_cache[layer_idx], key_states], dim=-2
                     )
@@ -1363,3 +1394,91 @@ class patched_SamMaskDecoder(torch.nn.Module):
         else:
             outputs = outputs + (None,)  # noqa: RUF005
         return outputs
+
+
+def rewrite_loop_for_square_mask(mask: torch.Tensor, seq: torch.Tensor):
+    """
+    Rewrites the loop in:
+
+    .. code-block:: python
+
+        attention_mask = torch.full(
+            [1, seq_length, seq_length], torch.finfo(q.dtype).min, dtype=q.dtype
+        )
+        for i in range(1, len(seq)):
+            attention_mask[..., seq[i - 1] : seq[i], seq[i - 1] : seq[i]] = 0
+    """
+    r = torch.arange(0, mask.shape[-1], dtype=torch.int64)
+    less0 = (r.reshape((-1, 1)) < seq.reshape((1, -1))).to(torch.int64)
+    less = less0.sum(axis=-1, keepdim=True) + 1
+    sq = less * less.T
+    look = (
+        torch.max(seq.min() == 0, less != less.max())
+        * torch.max(seq.max() == mask.shape[-1], less != less.min())
+        * less
+    )
+    filt = (sq != look**2).to(mask.dtype)
+    return mask * filt
+
+
+class patched_VisionAttention(torch.nn.Module):
+    _PATCHES_ = ["forward"]
+    _PATCHED_CLASS_ = transformers.models.qwen2_vl.modeling_qwen2_vl.VisionAttention
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        cu_seqlens: torch.Tensor,
+        rotary_pos_emb: Optional[torch.Tensor] = None,
+        position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
+    ) -> torch.Tensor:
+        seq_length = hidden_states.shape[0]
+        q, k, v = (
+            self.qkv(hidden_states)
+            .reshape(seq_length, 3, self.num_heads, -1)
+            .permute(1, 0, 2, 3)
+            .unbind(0)
+        )
+        if position_embeddings is None:
+            transformers.models.qwen2_vl.modeling_qwen2_vl.logger.warning_once(
+                "The attention layers in this model are transitioning from "
+                " computing the RoPE embeddings internally "
+                "through `rotary_pos_emb` (2D tensor of RoPE theta values), "
+                "to using externally computed "
+                "`position_embeddings` (Tuple of tensors, containing cos and sin)."
+                " In v4.54 `rotary_pos_emb` will be "
+                "removed and `position_embeddings` will be mandatory."
+            )
+            emb = torch.cat((rotary_pos_emb, rotary_pos_emb), dim=-1)
+            cos = emb.cos()
+            sin = emb.sin()
+        else:
+            cos, sin = position_embeddings
+        q, k = transformers.models.qwen2_vl.modeling_qwen2_vl.apply_rotary_pos_emb_vision(
+            q, k, cos, sin
+        )
+
+        attention_mask = torch.full(
+            [1, seq_length, seq_length],
+            torch.finfo(q.dtype).min,
+            device=q.device,
+            dtype=q.dtype,
+        )
+        # for i in range(1, len(cu_seqlens)):
+        #     attention_mask[..., cu_seqlens[i - 1] : cu_seqlens[i],
+        #                         cu_seqlens[i - 1] : cu_seqlens[i]] = 0
+        attention_mask = rewrite_loop_for_square_mask(attention_mask, cu_seqlens)
+
+        q = q.transpose(0, 1)
+        k = k.transpose(0, 1)
+        v = v.transpose(0, 1)
+        attn_weights = torch.matmul(q, k.transpose(1, 2)) / math.sqrt(self.head_dim)
+        attn_weights = attn_weights + attention_mask
+        attn_weights = torch.nn.functional.softmax(
+            attn_weights, dim=-1, dtype=torch.float32
+        ).to(q.dtype)
+        attn_output = torch.matmul(attn_weights, v)
+        attn_output = attn_output.transpose(0, 1)
+        attn_output = attn_output.reshape(seq_length, -1)
+        attn_output = self.proj(attn_output)
+        return attn_output

onnx_diagnostic/torch_export_patches/serialization/transformers_impl.py

@@ -261,7 +261,9 @@ def unflatten_encoder_decoder_cache(
 ) -> EncoderDecoderCache:
     """Restores a :class:`transformers.cache_utils.EncoderDecoderCache` from python objects."""
     dictionary = torch.utils._pytree._dict_unflatten(values, context)
-    return EncoderDecoderCache(**dictionary)
+    return EncoderDecoderCache(
+        dictionary["self_attention_cache"], dictionary["cross_attention_cache"]
+    )
 
 
 #############

onnx_diagnostic/torch_models/hghub/hub_data.py

@@ -37,6 +37,7 @@ __data_arch__ = textwrap.dedent(
     DebertaModel,feature-extraction
     DebertaV2Model,feature-extraction
     DecisionTransformerModel,reinforcement-learning
+    DeepseekV3ForCausalLM,text-generation
     DeiTModel,image-feature-extraction
     DetrModel,image-feature-extraction
     Dinov2Model,image-feature-extraction
@@ -52,9 +53,12 @@ __data_arch__ = textwrap.dedent(
     GPTJModel,feature-extraction
     GPTNeoModel,feature-extraction
     GPTNeoXForCausalLM,text-generation
+    GptOssForCausalLM,text-generation
     GemmaForCausalLM,text-generation
     Gemma2ForCausalLM,text-generation
     Gemma3ForConditionalGeneration,image-text-to-text
+    Gemma3ForCausalLM,text-generation
+    Glm4vMoeForConditionalGeneration,image-text-to-text
     GraniteForCausalLM,text-generation
     GroupViTModel,feature-extraction
     HieraForImageClassification,image-classification
@@ -107,6 +111,7 @@ __data_arch__ = textwrap.dedent(
     PvtForImageClassification,image-classification
     Qwen2ForCausalLM,text-generation
     Qwen2_5_VLForConditionalGeneration,image-text-to-text
+    Qwen3MoeForCausalLM,text-generation
     RTDetrForObjectDetection,object-detection
     RegNetModel,image-feature-extraction
     RemBertModel,feature-extraction

onnx_diagnostic/torch_models/validate.py

@@ -426,6 +426,7 @@ def validate_model(
             print(f"[validate_model] validate model id {model_id!r}, subfolder={subfolder!r}")
         else:
             print(f"[validate_model] validate model id {model_id!r}")
+        print(f"[validate_model] patch={patch!r}")
         if model_options:
             print(f"[validate_model] model_options={model_options!r}")
         print(f"[validate_model] get dummy inputs with input_options={input_options}...")

{onnx_diagnostic-0.7.6.dist-info → onnx_diagnostic-0.7.7.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: onnx-diagnostic
-Version: 0.7.6
+Version: 0.7.7
 Summary: Investigate ONNX models
 Home-page: https://github.com/sdpython/onnx-diagnostic
 Author: Xavier Dupré
@@ -27,7 +27,7 @@ Requires-Dist: numpy
 Requires-Dist: onnx>=1.16.0
 Requires-Dist: onnxruntime>=1.21
 Requires-Dist: optree
-Requires-Dist: torch>=2.7
+Requires-Dist: torch>=2.8
 Requires-Dist: torch_geometric
 Dynamic: author
 Dynamic: author-email

{onnx_diagnostic-0.7.6.dist-info → onnx_diagnostic-0.7.7.dist-info}/RECORD

@@ -1,22 +1,22 @@
-onnx_diagnostic/__init__.py,sha256=p8ROK5tw0ZrpyWOL_GKb_10U5K4JDHEGIZP4h9WHz_w,173
+onnx_diagnostic/__init__.py,sha256=JdBGK1KI9-G1MdBmNPIE7pLfaEHkMhgCq6h_YPJEYdw,173
 onnx_diagnostic/__main__.py,sha256=YmyV_Aq_ianDlHyKLHMa6h8YK3ZmFPpLVHLKjM91aCk,79
-onnx_diagnostic/_command_lines_parser.py,sha256=wU8L2C6f4BOrZMYM4WbHF6PDndeL7XKa__bx6BrKQs0,31227
+onnx_diagnostic/_command_lines_parser.py,sha256=8JlT1vzyGztkJT2v6lpQx5itLKY4FYlpFng3z8n3TAU,32937
 onnx_diagnostic/api.py,sha256=BhCl_yCd78N7TlVtPOHjeYv1QBEy39TjZ647rcHqLh0,345
 onnx_diagnostic/doc.py,sha256=t3RELgfooYnVMAi0JSpggWkQEgUsREz8NmRvn0TnLI8,2829
 onnx_diagnostic/ext_test_case.py,sha256=emfQGiQSz5FVDhyJ1Acsv_Tast7tWl426TjtpNqxDBU,43558
 onnx_diagnostic/export/__init__.py,sha256=yEIoWiOeTwBsDhyYt2fTKuhtA0Ya1J9u9ZzMTOTWaWs,101
-onnx_diagnostic/export/dynamic_shapes.py,sha256=HYf2OEi7PmRSj8uxMD-wbdVxxejkWXTPBAkxoFeM27A,40811
-onnx_diagnostic/export/shape_helper.py,sha256=EQXHRVxwGpHRYhx8Y44Crqs640pmaIuKSwW1KJOW0IU,7501
+onnx_diagnostic/export/dynamic_shapes.py,sha256=Go4_sIwiolCy_m1djQ3U_bX6C1EFw4al3x-ty-PsuME,41393
+onnx_diagnostic/export/shape_helper.py,sha256=PI_SgE1MNRKSrQ414eYoBZ54QGZbYisHSvqi9tstL2s,7795
 onnx_diagnostic/export/validate.py,sha256=_PGUql2DJhIgGKo0WjTGUc5AgsZUx8fEs00MePy-w98,6043
 onnx_diagnostic/helpers/__init__.py,sha256=GJ2GT7cgnlIveVUwMZhuvUwidbTJaKv8CsSIOpZDsJg,83
 onnx_diagnostic/helpers/_log_helper.py,sha256=zZ7AqGpiF3O2-9N8fLuPeB5VfJpbg3tQ20ccDRdQPVE,16753
 onnx_diagnostic/helpers/args_helper.py,sha256=SRWnqC7EENg09RZlA50B_PcdiIhdbgA4C3ACfzl5nMs,4419
 onnx_diagnostic/helpers/bench_run.py,sha256=CGA6VMJZMH2gDhVueT9ypNm4PMcjGrrGFYp08nhWj9k,16539
-onnx_diagnostic/helpers/cache_helper.py,sha256=0DBawuOi9dYXmKmgAyT7dcZxbEINQWVaAcD4mkK_0N8,21912
-onnx_diagnostic/helpers/config_helper.py,sha256=9h1NWC9RLmu43Yf5Cz9usjMdLiyLWXMhwgE4Lg-eOU8,3889
+onnx_diagnostic/helpers/cache_helper.py,sha256=sd9hnOW8uCU3yqvIB8tnxANRYEl1V_Ej8WDZMMI9VR8,24566
+onnx_diagnostic/helpers/config_helper.py,sha256=ZrwdQwG3atXzto1VLUzCVOzyBIWkih_EGc3qKHuluZw,4139
 onnx_diagnostic/helpers/doc_helper.py,sha256=pl5MZd3_FaE8BqQnqoBuSBxoNCFcd2OJd3eITUSku5c,5897
 onnx_diagnostic/helpers/graph_helper.py,sha256=hevQT5a7_QuriVPQcbT5qe18n99Doyl5h3-qshx1-uk,14093
-onnx_diagnostic/helpers/helper.py,sha256=Vz06pWFvkkTUHgHJQv2VXrAja1a6jqCT9Cz-_EKx6Ec,62757
+onnx_diagnostic/helpers/helper.py,sha256=OsQz2um10DgGiX3fgOulTDFQop0wCMX6shPonQgN71w,62940
 onnx_diagnostic/helpers/log_helper.py,sha256=rBYtZo85n61uQRIpxKpxkKTtKv-bDZvAc1J1uHRDosc,82567
 onnx_diagnostic/helpers/memory_peak.py,sha256=OT6mz0muBbBZY0pjgW2_eCk_lOtFRo-5w4jFo2Z6Kok,6380
 onnx_diagnostic/helpers/mini_onnx_builder.py,sha256=FgK-Kws1WpSYdYJCPyONwQYY3AjbgUHimZlaYyiNUfE,21286
@@ -36,7 +36,7 @@ onnx_diagnostic/reference/ops/op_add_add_mul_mul.py,sha256=CXQVtgVrT066gDJFwxL4n
 onnx_diagnostic/reference/ops/op_attention.py,sha256=ThALMDF53v3QeG1bohi0bvX2o90HZhGJbbAFOtwEHPE,2027
 onnx_diagnostic/reference/ops/op_average_pool_grad.py,sha256=zMcOtjB7hWySfIIFXogcmj3xxCWwxEX_g2VKg-SOAEs,2360
 onnx_diagnostic/reference/ops/op_bias_softmax.py,sha256=dcXsw2chxc8-puIkI0LFsBxKOJaCSovDcF1HkgboQp0,524
-onnx_diagnostic/reference/ops/op_cast_like.py,sha256=tH2pYFPluHK9r_cdkqnk9k2q3dPzaeKgQZt_Pjw9N4o,1383
+onnx_diagnostic/reference/ops/op_cast_like.py,sha256=Sn0QA47bQ8laXJUl7VaxtsWCzCTpwkAQdv7F4A-AB5c,1557
 onnx_diagnostic/reference/ops/op_complex.py,sha256=OobDrRNYcktdCdTJzOQBesrKC8vsKxuHIi7Yev1DJrs,651
 onnx_diagnostic/reference/ops/op_concat.py,sha256=seW71-QDKzv9QQhjhjThKip0Y3d9nkVd7Hs1A2nNQjk,519
 onnx_diagnostic/reference/ops/op_constant_of_shape.py,sha256=3G9TRxaUoqYudnKrVHBEblo_16qMl0c9wNZqnQeNJJ4,2009
@@ -60,7 +60,7 @@ onnx_diagnostic/reference/ops/op_skip_layer_normalization.py,sha256=oJ7fQNx2iQh9
 onnx_diagnostic/reference/ops/op_slice.py,sha256=yRxfYBs8b7QezyyG9JHCD8MIJHij2qR2NNDpBmD3FJI,705
 onnx_diagnostic/reference/ops/op_transpose_cast.py,sha256=ifef74rvh0Yvq1Zx51B4mfnISbxV9uRg9DFjkdL1_68,361
 onnx_diagnostic/reference/ops/op_tri_matrix.py,sha256=Yn2gxAyygcwtF5Hjau9ihXDAzul0BAkdqVimVahtFBU,519
-onnx_diagnostic/reference/torch_ops/__init__.py,sha256=eZ7FBYH9Ta3BC0f7BJQAnIqiebqRkCt2_T3ktpVV6iQ,1150
+onnx_diagnostic/reference/torch_ops/__init__.py,sha256=fIDMG0KhNYuB4q_57Z5zz0ME6_w8Q0tnqUCooijoOW4,1163
 onnx_diagnostic/reference/torch_ops/_op_run.py,sha256=EEUIwfbRldEFDhULquYhk9x5ZDa9t6f2mKJ1sM__D6A,10517
 onnx_diagnostic/reference/torch_ops/access_ops.py,sha256=Zfs5OF03PV1CqlCqKI5VV-c4MY3KyQxmO7QZksxQjX8,3274
 onnx_diagnostic/reference/torch_ops/binary_ops.py,sha256=-KxMcCYGDTcZyOss9qU1nU0rmdyg9SdVHJQohseSTcQ,2653
@@ -71,44 +71,44 @@ onnx_diagnostic/reference/torch_ops/other_ops.py,sha256=FnCY60mhdrzrsiHgvN-XpFRH
 onnx_diagnostic/reference/torch_ops/reduce_ops.py,sha256=9gFfraPTQbe_ZEUNCUis1JSmA5dj4tSzjAOpZPJKG4Y,5102
 onnx_diagnostic/reference/torch_ops/sequence_ops.py,sha256=3EiVKpGfN4d1Iry4hgnr3MIJyEEKUrAIDgmRGsUXXa0,2297
 onnx_diagnostic/reference/torch_ops/shape_ops.py,sha256=pJrNR2UB4PlWl6cv4EDl1uGl8YTBUUMQkhJcsh5K4sA,4291
-onnx_diagnostic/reference/torch_ops/unary_ops.py,sha256=E8Ys1eZsOTsucBKoXb1_Kl5LbBDygniDvW2BvN4IPMo,1708
+onnx_diagnostic/reference/torch_ops/unary_ops.py,sha256=dwu6HPr4V_roxu85U3VLTtDLx5bfxKalT_-zlQxZ5wc,1850
 onnx_diagnostic/tasks/__init__.py,sha256=uWFP7HIr-VnxmXD5i_QAfXnLXc1HwUq2e8v9cKLqraQ,2492
-onnx_diagnostic/tasks/automatic_speech_recognition.py,sha256=H94rxeiAjcJdECA1g95G_U9fZfpXk6dfjNKKYuvc4Qc,7130
-onnx_diagnostic/tasks/feature_extraction.py,sha256=4M5bE6lGf7KG9qnlim9EzVmqptL17jsSnWgT-gYtcYg,5534
-onnx_diagnostic/tasks/fill_mask.py,sha256=Rvrz0j_DQu-vf4CSSAZMBMXb2EuHvOCzRZwj8Cy8yfg,2620
-onnx_diagnostic/tasks/image_classification.py,sha256=x1XfeWAOe0r_s9kU6WENoYxjfoRTp1pkwKgIveoLbUw,4627
-onnx_diagnostic/tasks/image_text_to_text.py,sha256=xAp3RcFGwVYdByDqAlTvO82mPZiwiFhVQMWL1yV7ZZU,16224
-onnx_diagnostic/tasks/mask_generation.py,sha256=UQExMner3As8KznK-y4MjycYeUz5POcVR-oLEiAXmK4,5002
-onnx_diagnostic/tasks/mixture_of_expert.py,sha256=DgIsbwzV4smysOK83wny91k3ix1Qt2tSFXLGLoz4WOo,2796
-onnx_diagnostic/tasks/object_detection.py,sha256=xRBH9JZxBQf0SVSTJP6d-VVCKqrw7JAeif1joHfiYng,4224
-onnx_diagnostic/tasks/sentence_similarity.py,sha256=soL6QxLvyjtQ-3tQ3nCFxrcrk_4a8tuAjil8zYQ_pXk,2635
-onnx_diagnostic/tasks/summarization.py,sha256=LZ8A8wl6cd8kWSc6k5vLHa_XZkm35rYkTRv8iUYtr6I,8268
-onnx_diagnostic/tasks/text2text_generation.py,sha256=SzmgECK05H_fieIa2rlY-MQUmvNRotOo05J2eiMwjIM,8578
-onnx_diagnostic/tasks/text_classification.py,sha256=dO_LLbwwv0OJfIa9DqxQqAGUDuz3iIF1XkafzaYJdJw,2691
-onnx_diagnostic/tasks/text_generation.py,sha256=xudXQlCC-chd3p_T1k5qmvLE698wIWoVLAUgAsJ8GaQ,13257
+onnx_diagnostic/tasks/automatic_speech_recognition.py,sha256=tguoQO77okXo8vcJrN2FAmpO9zkq04WSY8OKgm5sqRw,7185
+onnx_diagnostic/tasks/feature_extraction.py,sha256=pcFON5uGKoykjg52bMsvpYG7KJvXd8JDC43rAjXIzB0,5572
+onnx_diagnostic/tasks/fill_mask.py,sha256=Z0OyDs3pcnjJLzZBbS52d6pa6jh6m2Uy8-h3nF5wbDA,2675
+onnx_diagnostic/tasks/image_classification.py,sha256=nLpBBB1Gkog3Fk6pu2waiHcuQr4ILPptc9FhQ-pn460,4682
+onnx_diagnostic/tasks/image_text_to_text.py,sha256=syi7MzOKnHlWfqsfQsbmY6zcwtj8flg5_jVqodmIKcY,16806
+onnx_diagnostic/tasks/mask_generation.py,sha256=fjdD3rd-O-mFL0hQy3la3JXKth_0bH2HL7Eelq-3Dbs,5057
+onnx_diagnostic/tasks/mixture_of_expert.py,sha256=al4tk1BrHidtRiHlAaiflWiJaAte0d5M8WcBioANG9k,2808
+onnx_diagnostic/tasks/object_detection.py,sha256=3FiT8ya5FCd9lwjQCRXhAwXspNwYTlAD3Gpk8aAcG5w,4279
+onnx_diagnostic/tasks/sentence_similarity.py,sha256=Azpw9hxLwxM7O14Rj0mzH73LrrYGhiBm8fIb_tVnXCM,2690
+onnx_diagnostic/tasks/summarization.py,sha256=tH2aX8undH8n-CGzDp-d-nBil2TD4Kv3PqeiaiFLkqY,8327
+onnx_diagnostic/tasks/text2text_generation.py,sha256=-5Iy938AsQBXmA2chxYc7MUZQ7ZlmWJQ7k_O66wGTTo,8637
+onnx_diagnostic/tasks/text_classification.py,sha256=JbKjnMoocr-k3-RVC5b-68kZaKJTPTm3aKSTXShHEo0,2746
+onnx_diagnostic/tasks/text_generation.py,sha256=i1tX_cOMnKwIvEcm2EFNWtZ4DwJGcQ6EGQGSJmu69Oc,12858
 onnx_diagnostic/tasks/text_to_image.py,sha256=mOS3Ruosi3hzRMxXLDN7ZkAbi7NnQb7MWwQP_okGVHs,2962
 onnx_diagnostic/tasks/zero_shot_image_classification.py,sha256=1iqYamkq5kZNXEXsySw748ernc0O94GkwpYAIEl6Kj4,4659
 onnx_diagnostic/torch_export_patches/__init__.py,sha256=0SaZedwznm1hQUCvXZsGZORV5vby954wEExr5faepGg,720
-onnx_diagnostic/torch_export_patches/onnx_export_errors.py,sha256=NVJNhwr5LVwKzU9RsR2b8fOounmQ1W3nx_k-6XirASc,23701
+onnx_diagnostic/torch_export_patches/onnx_export_errors.py,sha256=Nx3HLII-KIemfMydraTRlwK9O0kgVug57SiLT9y9KOY,23749
 onnx_diagnostic/torch_export_patches/onnx_export_serialization.py,sha256=klvqiMjccwGhiRnLRVbwTi5WWkMfvtnOV5ycirPcAdA,11354
 onnx_diagnostic/torch_export_patches/patch_expressions.py,sha256=vr4tt61cbDnaaaduzMj4UBZ8OUtr6GfDpIWwOYqjWzs,3213
-onnx_diagnostic/torch_export_patches/patch_inputs.py,sha256=_z6hcabYRGJ6VYiHuMzxYyFF8pYW-Xwcm60Wst3O8zA,7779
+onnx_diagnostic/torch_export_patches/patch_inputs.py,sha256=3ySY1nAzINSS1hAzTycwfdbPas8G5CDL2MjnaAHBkMU,7825
 onnx_diagnostic/torch_export_patches/patch_module.py,sha256=R2d9IHM-RwsBKDsxuBIJnEqMoxbS9gd4YWFGG2wwV5A,39881
 onnx_diagnostic/torch_export_patches/patch_module_helper.py,sha256=2U0AdyZuU0W54QTdE7tY7imVzMnpQ5091ADNtTCkT8Y,6967
 onnx_diagnostic/torch_export_patches/eval/__init__.py,sha256=57x62uZNA80XiWgkG8Fe0_8YJcIVrvKLPqvwLDPJwgc,24008
 onnx_diagnostic/torch_export_patches/eval/model_cases.py,sha256=DTvdHPtNQh25Akv5o3D4Jxf1L1-SJ7w14tgvj8AAns8,26577
 onnx_diagnostic/torch_export_patches/patches/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-onnx_diagnostic/torch_export_patches/patches/patch_torch.py,sha256=KaZ8TjDa9ATgT4HllYzzoNf_51q_yOj_GuF5NYjPCrU,18913
-onnx_diagnostic/torch_export_patches/patches/patch_transformers.py,sha256=ZmBCpkHx7FV8eIQDXavxjK4PZLSOn5ApT4bVQbJIqMg,56770
+onnx_diagnostic/torch_export_patches/patches/patch_torch.py,sha256=g1UjL6A6iB7Qh2Cs1efuKk5377IvsSnZXUk3jNeRu_E,18830
+onnx_diagnostic/torch_export_patches/patches/patch_transformers.py,sha256=3pAm3cPDq_HxpKCJNLLyrO53ZAnByjg7uKzpUPhz7nc,61378
 onnx_diagnostic/torch_export_patches/serialization/__init__.py,sha256=BHLdRPtNAtNPAS-bPKEj3-foGSPvwAbZXrHzGGPDLEw,1876
 onnx_diagnostic/torch_export_patches/serialization/diffusers_impl.py,sha256=drq3EH_yjcSuIWYsVeUWm8Cx6YCZFU6bP_1PLtPfY5I,945
-onnx_diagnostic/torch_export_patches/serialization/transformers_impl.py,sha256=jcSO2pvzbLA3JXVl3uDfoak7gM3J1SnmxshdSFxzE9w,9280
+onnx_diagnostic/torch_export_patches/serialization/transformers_impl.py,sha256=dAKi4zujlBxDvxvaVI_qH4qW9AlpVFMtCkvGTNCJCUY,9353
 onnx_diagnostic/torch_models/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 onnx_diagnostic/torch_models/llms.py,sha256=soyg4yC87ptGoeulJhKqw5opGmuLvH1pn_ZDXZ4Jr8E,90
-onnx_diagnostic/torch_models/validate.py,sha256=FNqi-Dg7-S_dSQDdcUjc5ERgTquiXclwGv0UZxVHYbY,64866
+onnx_diagnostic/torch_models/validate.py,sha256=hMe4fbRDubSLfWc3XguRl_fjqQBa7-2zVKxhge9iaqc,64917
 onnx_diagnostic/torch_models/hghub/__init__.py,sha256=vi1Q7YHdddj1soiBN42MSvJdFqe2_KUoWafHISjwOu8,58
 onnx_diagnostic/torch_models/hghub/hub_api.py,sha256=Bvr-sTAhS6s6UCkt-KsY_7Mdai08-AQzvHrzbYCSuvk,13186
-onnx_diagnostic/torch_models/hghub/hub_data.py,sha256=NTTDsCtIVvYnr5J3rlcq0GSGDOzOPzq9Tsnb3oVf4q8,8309
+onnx_diagnostic/torch_models/hghub/hub_data.py,sha256=W05mciqUqhaYEfYNHtUeuwOMOZoQTuDidRLEIx4z1CE,8523
 onnx_diagnostic/torch_models/hghub/hub_data_cached_configs.py,sha256=XZ_PsMUmMooJw5pBqEtDMOWbbLYxDcJdRWf-FNz2cYg,279674
 onnx_diagnostic/torch_models/hghub/model_inputs.py,sha256=SDRLCA2zivEHIKr2RRRP-dZNiNUcpYS3EgP0unLExxY,11046
 onnx_diagnostic/torch_models/untrained/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -117,8 +117,8 @@ onnx_diagnostic/torch_models/untrained/llm_tiny_llm.py,sha256=QXw_Bs2SzfeiQMf-tm
 onnx_diagnostic/torch_onnx/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 onnx_diagnostic/torch_onnx/runtime_info.py,sha256=1g9F_Jf9AAgYQU4stbsrFXwQl-30mWlQrFbQ7val8Ps,9268
 onnx_diagnostic/torch_onnx/sbs.py,sha256=1EL25DeYFzlBSiFG_XjePBLvsiItRXbdDrr5-QZW2mA,16878
-onnx_diagnostic-0.7.6.dist-info/licenses/LICENSE.txt,sha256=Vv6TXglX6Rc0d-f8aREhayhT-6PMQXEyOmI2NKlUCMc,1045
-onnx_diagnostic-0.7.6.dist-info/METADATA,sha256=dPByK3aJs_ydnyy_Vfzlavdm00um0Ad3V6pu-78dWw0,7431
-onnx_diagnostic-0.7.6.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-onnx_diagnostic-0.7.6.dist-info/top_level.txt,sha256=KwNkXewmcobM3ZT1DJLVWH6ebJzA5qKg7cWqKfpGNT4,16
-onnx_diagnostic-0.7.6.dist-info/RECORD,,
+onnx_diagnostic-0.7.7.dist-info/licenses/LICENSE.txt,sha256=Vv6TXglX6Rc0d-f8aREhayhT-6PMQXEyOmI2NKlUCMc,1045
+onnx_diagnostic-0.7.7.dist-info/METADATA,sha256=zus-hHQ_QtzpPCf--m_YA8MnWvfb8rmwbCGEy-ZlLro,7431
+onnx_diagnostic-0.7.7.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+onnx_diagnostic-0.7.7.dist-info/top_level.txt,sha256=KwNkXewmcobM3ZT1DJLVWH6ebJzA5qKg7cWqKfpGNT4,16
+onnx_diagnostic-0.7.7.dist-info/RECORD,,