onnx-diagnostic 0.7.4__py3-none-any.whl → 0.7.5__py3-none-any.whl

onnx_diagnostic/tasks/feature_extraction.py

@@ -1,17 +1,15 @@
 from typing import Any, Callable, Dict, Optional, Tuple
 import torch
 from ..helpers.config_helper import update_config, check_hasattr
+from ..helpers.cache_helper import make_dynamic_cache, make_encoder_decoder_cache
 
 __TASK__ = "feature-extraction"
 
 
 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_attention_heads=min(config.num_attention_heads, 4),
-    )
+    check_hasattr(config, "num_hidden_layers")
+    kwargs = dict(num_hidden_layers=min(config.num_hidden_layers, 2))
     update_config(config, kwargs)
     return kwargs
 
@@ -22,6 +20,12 @@ def get_inputs(
     batch_size: int,
     sequence_length: int,
     dummy_max_token_id: int,
+    sequence_length2: int = 3,
+    decoder_attention_heads: Optional[int] = None,
+    encoder_attention_heads: Optional[int] = None,
+    encoder_ffn_dim: Optional[int] = None,
+    decoder_ffn_dim: Optional[int] = None,
+    num_hidden_layers: Optional[int] = None,
     add_second_input: int = 1,
     **kwargs,  # unused
 ):
@@ -50,6 +54,66 @@ def get_inputs(
         ),
         attention_mask=torch.ones((batch_size, sequence_length)).to(torch.int64),
     )
+    if (
+        encoder_attention_heads
+        and decoder_attention_heads
+        and encoder_ffn_dim
+        and decoder_ffn_dim
+        and num_hidden_layers
+    ):
+        inputs["past_key_values"] = make_encoder_decoder_cache(
+            make_dynamic_cache(
+                [
+                    (
+                        torch.randn(
+                            batch_size,
+                            encoder_attention_heads,
+                            sequence_length,
+                            encoder_ffn_dim,
+                        ),
+                        torch.randn(
+                            batch_size,
+                            encoder_attention_heads,
+                            sequence_length,
+                            encoder_ffn_dim,
+                        ),
+                    )
+                    for i in range(num_hidden_layers)
+                ]
+            ),
+            make_dynamic_cache(
+                [
+                    (
+                        torch.randn(
+                            batch_size,
+                            decoder_attention_heads,
+                            sequence_length2,
+                            decoder_ffn_dim,
+                        ),
+                        torch.randn(
+                            batch_size,
+                            decoder_attention_heads,
+                            sequence_length2,
+                            decoder_ffn_dim,
+                        ),
+                    )
+                    for i in range(num_hidden_layers)
+                ]
+            ),
+        )
+        cache_length = "cache_length_key"
+        cache_length2 = "cache_length_val"
+        shapes["past_key_values"] = [  # type: ignore[assignment]
+            [
+                [{0: batch, 2: cache_length} for _ in range(num_hidden_layers)],
+                [{0: batch, 2: cache_length} for _ in range(num_hidden_layers)],
+            ],
+            [
+                [{0: batch, 2: cache_length2} for _ in range(num_hidden_layers)],
+                [{0: batch, 2: cache_length2} for _ in range(num_hidden_layers)],
+            ],
+        ]
+
     res = dict(inputs=inputs, dynamic_shapes=shapes)
     if add_second_input:
         assert (
@@ -61,6 +125,12 @@ def get_inputs(
             batch_size=batch_size + 1,
             sequence_length=sequence_length + add_second_input,
             dummy_max_token_id=dummy_max_token_id,
+            sequence_length2=sequence_length2,
+            decoder_attention_heads=decoder_attention_heads,
+            encoder_attention_heads=encoder_attention_heads,
+            encoder_ffn_dim=encoder_ffn_dim,
+            decoder_ffn_dim=decoder_ffn_dim,
+            num_hidden_layers=num_hidden_layers,
             add_second_input=0,
             **kwargs,
         )["inputs"]
@@ -80,4 +150,15 @@ def random_input_kwargs(config: Any) -> Tuple[Dict[str, Any], Callable]:
         sequence_length=30,
         dummy_max_token_id=31999 if config is None else (config.vocab_size - 1),
     )
+    for att in [
+        "decoder_attention_heads",
+        "encoder_attention_heads",
+        "encoder_ffn_dim",
+        "decoder_ffn_dim",
+        "num_hidden_layers",
+    ]:
+        if hasattr(config, att):
+            kwargs[att] = getattr(config, att)
+    kwargs["decoder_ffn_dim"] = kwargs["encoder_ffn_dim"] = 64
+    print(kwargs)
     return kwargs, get_inputs

onnx_diagnostic/tasks/text2text_generation.py

@@ -69,8 +69,8 @@ def get_inputs(
     ), f"Not yet implemented for cls_cache={kwargs['cls_cache']!r}."
     batch = torch.export.Dim("batch", min=1, max=1024)
     seq_length = "seq_length"  # torch.export.Dim("seq_length", min=1, max=4096)
-    cache_length = "cache_length_key"  # torch.export.Dim("cache_length", min=1, max=4096)
-    cache_length2 = "cache_length_val"  # torch.export.Dim("cache_length2", min=1, max=4096)
+    cache_length = "cache_length_key"
+    cache_length2 = "cache_length_val"
 
     shapes = {
         "input_ids": {0: batch, 1: seq_length},

onnx_diagnostic/torch_export_patches/onnx_export_errors.py

@@ -16,6 +16,8 @@ def get_function(name: str) -> Tuple[type, Callable]:
     module_name = ".".join(spl[:-1])
     fname = spl[-1]
     mod = importlib.import_module(module_name)
+    if not hasattr(mod, fname):
+        return None, None
     return mod, getattr(mod, fname)
 
 
@@ -33,12 +35,16 @@ def get_patches(mod, verbose: int = 0) -> Tuple[str, List[Any]]:
                 doc = v.__doc__.lstrip()
                 if doc.startswith("manual patch"):
                     continue
-                reg = re.compile("[[]patch:([a-z_A-Z.]+)[]]")
+                reg = re.compile("[\\[]patch:([a-z_A-Z.]+)[\\]]")
                 fall = reg.findall(doc)
                 assert (
                     len(fall) == 1
                 ), f"Unable to find patching information for {v} in \n{doc}"
                 fmod, f = get_function(fall[0])
+                if fmod is None and f is None:
+                    # The function does not exist in this version of transformers.
+                    # No patch is needed.
+                    continue
                 to_patch.append({"module": fmod, "function": f, "patch": v})
 
     name = mod.__name__

onnx_diagnostic/torch_export_patches/patches/patch_transformers.py

@@ -255,7 +255,8 @@ class patched_DynamicCache:
         """
         # Update the number of seen tokens
         if layer_idx == 0:
-            self._seen_tokens += key_states.shape[-2]
+            if hasattr(self, "_seen_tokens"):
+                self._seen_tokens += key_states.shape[-2]
 
         # Update the cache
         if key_states is not None:
@@ -294,7 +295,8 @@ class patched_DynamicCache:
         if self.get_seq_length() <= max_length:
             return
 
-        self._seen_tokens = max_length
+        if hasattr(self, "_seen_tokens"):
+            self._seen_tokens = max_length
         for idx in range(len(self.key_cache)):
             if self.key_cache[idx].numel():
                 self.key_cache[idx] = self.key_cache[idx][..., :max_length, :]
@@ -862,6 +864,91 @@ def patched_dynamic_rope_update(rope_forward):
     return wrapper
 
 
+def common_eager_attention_forward(
+    module: torch.nn.Module,
+    query: torch.Tensor,
+    key: torch.Tensor,
+    value: torch.Tensor,
+    attention_mask: Optional[torch.Tensor],
+    scaling: Optional[float] = None,
+    dropout: float = 0.0,
+    head_mask: Optional[torch.Tensor] = None,
+    **kwargs,
+):
+    if scaling is None:
+        scaling = query.size(-1) ** -0.5
+
+    attn_weights = torch.matmul(query, key.transpose(2, 3)) * scaling
+    if attention_mask is not None:
+        # The two following lines were added.
+        if attention_mask is not None and attention_mask.ndim == 4:
+            attention_mask = attention_mask[:, :, :, : key.shape[-2]]
+        attn_weights = attn_weights + attention_mask
+
+    attn_weights = torch.nn.functional.softmax(attn_weights, dim=-1)
+
+    if head_mask is not None:
+        attn_weights = attn_weights * head_mask.view(1, -1, 1, 1)
+
+    attn_weights = torch.nn.functional.dropout(
+        attn_weights, p=dropout, training=module.training
+    )
+    attn_output = torch.matmul(attn_weights, value)
+    attn_output = attn_output.transpose(1, 2).contiguous()
+
+    return attn_output, attn_weights
+
+
+def patched_model_bart_eager_attention_forward(
+    module: torch.nn.Module,
+    query: torch.Tensor,
+    key: torch.Tensor,
+    value: torch.Tensor,
+    attention_mask: Optional[torch.Tensor],
+    scaling: Optional[float] = None,
+    dropout: float = 0.0,
+    head_mask: Optional[torch.Tensor] = None,
+    **kwargs,
+):
+    """[patch:transformers.models.bart.modeling_bart.eager_attention_forward]"""
+    return common_eager_attention_forward(
+        module,
+        query,
+        key,
+        value,
+        attention_mask=attention_mask,
+        scaling=scaling,
+        dropout=dropout,
+        head_mask=head_mask,
+        **kwargs,
+    )
+
+
+def patched_modeling_marian_eager_attention_forward(
+    module: torch.nn.Module,
+    query: torch.Tensor,
+    key: torch.Tensor,
+    value: torch.Tensor,
+    attention_mask: Optional[torch.Tensor],
+    scaling: Optional[float] = None,
+    dropout: float = 0.0,
+    head_mask: Optional[torch.Tensor] = None,
+    **kwargs,
+):
+    """[patch:transformers.models.marian.modeling_marian.eager_attention_forward]"""
+    return common_eager_attention_forward(
+        module,
+        query,
+        key,
+        value,
+        attention_mask=attention_mask,
+        scaling=scaling,
+        dropout=dropout,
+        head_mask=head_mask,
+        **kwargs,
+    )
+
+
 class common_RotaryEmbedding(torch.nn.Module):
     @torch.no_grad()
     @patched_dynamic_rope_update
@@ -1093,4 +1180,6 @@ class patched_IdeficsAttention(torch.nn.Module):
         if output_attentions:
             attn_weights = None
 
-        return attn_output, attn_weights, past_key_value
+        if pv.Version(transformers.__version__) < pv.Version("4.53.99"):
+            return attn_output, attn_weights, past_key_value
+        return attn_output, attn_weights

onnx_diagnostic/torch_models/hghub/model_inputs.py

@@ -144,6 +144,11 @@ def get_untrained_model_with_inputs(
                 f"[get_untrained_model_with_inputs] config._attn_implementation="
                 f"{config._attn_implementation!r}"  # type: ignore[union-attr]
             )
+    elif verbose:
+        print(
+            f"[get_untrained_model_with_inputs] default config._attn_implementation="
+            f"{getattr(config, '_attn_implementation', '?')!r}"  # type: ignore[union-attr]
+        )
 
     if type(config) is dict and "_diffusers_version" in config:
         import diffusers

onnx_diagnostic/torch_models/validate.py

@@ -288,6 +288,7 @@ def validate_model(
     repeat: int = 1,
     warmup: int = 0,
     inputs2: int = 1,
+    output_names: Optional[List[str]] = None,
 ) -> Tuple[Dict[str, Union[int, float, str]], Dict[str, Any]]:
     """
     Validates a model.
@@ -338,6 +339,7 @@ def validate_model(
     :param inputs2: checks that the second set of inputs is reunning as well,
         this ensures that the model does support dynamism, the value is used
         as an increment to the first set of values (added to dimensions)
+    :param output_names: output names the onnx exporter should use
     :return: two dictionaries, one with some metrics,
         another one with whatever the function produces
 
@@ -433,6 +435,7 @@ def validate_model(
         )
         print(f"[validate_model] exporter={exporter!r}, optimization={optimization!r}")
         print(f"[validate_model] dump_folder={dump_folder!r}")
+        print(f"[validate_model] output_names={output_names}")
         summary["model_id"] = model_id
         summary["model_subfolder"] = subfolder or ""
 
@@ -631,6 +634,7 @@ def validate_model(
                     optimization=optimization,
                     do_run=do_run,
                     dump_folder=dump_folder,
+                    output_names=output_names,
                 )
         else:
             data["inputs_export"] = data["inputs"]
@@ -643,6 +647,7 @@ def validate_model(
                 optimization=optimization,
                 do_run=do_run,
                 dump_folder=dump_folder,
+                output_names=output_names,
             )
         summary.update(summary_export)
 
@@ -868,6 +873,7 @@ def call_exporter(
     optimization: Optional[str] = None,
     do_run: bool = False,
     dump_folder: Optional[str] = None,
+    output_names: Optional[List[str]] = None,
 ) -> Tuple[Dict[str, Union[int, float, str]], Dict[str, Any]]:
     """
     Calls an exporter on a model;
@@ -880,6 +886,7 @@ def call_exporter(
     :param optimization: optimization to do
     :param do_run: runs and compute discrepancies
     :param dump_folder: to dump additional information
+    :param output_names: list of output names to use with the onnx exporter
     :return: two dictionaries, one with some metrics,
         another one with whatever the function produces
     """
@@ -902,6 +909,7 @@ def call_exporter(
             quiet=quiet,
             verbose=verbose,
             optimization=optimization,
+            output_names=output_names,
         )
         return summary, data
     if exporter == "custom" or exporter.startswith("custom"):
@@ -913,6 +921,7 @@ def call_exporter(
             verbose=verbose,
             optimization=optimization,
             dump_folder=dump_folder,
+            output_names=output_names,
         )
         return summary, data
     if exporter == "modelbuilder":
@@ -923,6 +932,7 @@ def call_exporter(
             quiet=quiet,
             verbose=verbose,
             optimization=optimization,
+            output_names=output_names,
         )
         return summary, data
     raise NotImplementedError(
@@ -1090,7 +1100,7 @@ def validate_onnx_model(
     """
     import onnxruntime
 
-    def _mk(key):
+    def _mk(key, flavour=flavour):
         return f"{key}_{flavour}" if flavour else key
 
     summary: Dict[str, Any] = {}
@@ -1145,7 +1155,7 @@ def validate_onnx_model(
     )
     sess = _quiet_or_not_quiet(
         quiet,
-        _mk("onnx_ort_create"),
+        _mk("create_onnx_ort"),
         summary,
         data,
         (lambda source=source, providers=providers: cls_runtime(source, providers)),
@@ -1180,7 +1190,7 @@ def validate_onnx_model(
 
         got = _quiet_or_not_quiet(
             quiet,
-            _mk(f"time_onnx_ort_run{suffix}"),
+            _mk(f"run_onnx_ort{suffix}"),
             summary,
             data,
             (lambda sess=sess, feeds=feeds: sess.run(None, feeds)),
@@ -1211,6 +1221,7 @@ def call_torch_export_onnx(
     quiet: bool = False,
     verbose: int = 0,
     optimization: Optional[str] = None,
+    output_names: Optional[List[str]] = None,
 ) -> Tuple[Dict[str, Any], Dict[str, Any]]:
     """
     Exports a model into onnx.
@@ -1222,6 +1233,7 @@ def call_torch_export_onnx(
     :param quiet: catch exception or not
     :param verbose: verbosity
     :param optimization: optimization to do
+    :param output_names: output names to use
     :return: two dictionaries, one with some metrics,
         another one with whatever the function produces
     """
@@ -1276,6 +1288,8 @@ def call_torch_export_onnx(
             print("[call_torch_export_onnx] dynamo=False so...")
             print(f"[call_torch_export_onnx] args={string_type(args, with_shape=True)}")
             print(f"[call_torch_export_onnx] kwargs={string_type(kwargs, with_shape=True)}")
+    if output_names:
+        export_export_kwargs["output_names"] = output_names
     if opset:
         export_export_kwargs["opset_version"] = opset
     if verbose:
@@ -1346,6 +1360,7 @@ def call_torch_export_model_builder(
     quiet: bool = False,
     verbose: int = 0,
     optimization: Optional[str] = None,
+    output_names: Optional[List[str]] = None,
 ) -> Tuple[Dict[str, Any], Dict[str, Any]]:
     """
     Exports a model into onnx with :epkg:`ModelBuilder`.
@@ -1356,6 +1371,7 @@ def call_torch_export_model_builder(
     :param quiet: catch exception or not
     :param verbose: verbosity
     :param optimization: optimization to do
+    :param output_names: list of output names to use
     :return: two dictionaries, one with some metrics,
         another one with whatever the function produces
     """
@@ -1369,6 +1385,9 @@ def call_torch_export_model_builder(
     provider = data.get("model_device", "cpu")
     dump_folder = data.get("model_dump_folder", "")
     assert dump_folder, "dump_folder cannot be empty with ModelBuilder"
+    assert (
+        not output_names
+    ), f"output_names not empty, not supported yet, output_names={output_names}"
     cache_dir = os.path.join(dump_folder, "cache_mb")
     if not os.path.exists(cache_dir):
         os.makedirs(cache_dir)
@@ -1408,6 +1427,7 @@ def call_torch_export_custom(
     verbose: int = 0,
     optimization: Optional[str] = None,
     dump_folder: Optional[str] = None,
+    output_names: Optional[List[str]] = None,
 ) -> Tuple[Dict[str, Any], Dict[str, Any]]:
     """
     Exports a model into onnx.
@@ -1420,6 +1440,7 @@ def call_torch_export_custom(
     :param verbose: verbosity
     :param optimization: optimization to do
     :param dump_folder: to store additional information
+    :param output_names: list of output names to use
     :return: two dictionaries, one with some metrics,
         another one with whatever the function produces
     """
@@ -1504,6 +1525,8 @@ def call_torch_export_custom(
     )
     if opset:
         kws["target_opset"] = opset
+    if output_names:
+        kws["output_names"] = output_names
 
     epo, opt_stats = _quiet_or_not_quiet(
         quiet,

{onnx_diagnostic-0.7.4.dist-info → onnx_diagnostic-0.7.5.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: onnx-diagnostic
-Version: 0.7.4
+Version: 0.7.5
 Summary: Investigate ONNX models
 Home-page: https://github.com/sdpython/onnx-diagnostic
 Author: Xavier Dupré

{onnx_diagnostic-0.7.4.dist-info → onnx_diagnostic-0.7.5.dist-info}/RECORD

@@ -1,6 +1,6 @@
-onnx_diagnostic/__init__.py,sha256=dmZNMpFkDRd7ZCC6bC2fEFqvAhHhsqqua8ZE5LbOC9s,173
+onnx_diagnostic/__init__.py,sha256=759u8m5f7RZRqqatZb98hD3h2h5IG0CCVPhEq4f80aA,173
 onnx_diagnostic/__main__.py,sha256=YmyV_Aq_ianDlHyKLHMa6h8YK3ZmFPpLVHLKjM91aCk,79
-onnx_diagnostic/_command_lines_parser.py,sha256=65oUjJ2tgPxQgIKgOAI04jhOFRnGUSNivUNDVMZ-urU,28597
+onnx_diagnostic/_command_lines_parser.py,sha256=wU8L2C6f4BOrZMYM4WbHF6PDndeL7XKa__bx6BrKQs0,31227
 onnx_diagnostic/api.py,sha256=BhCl_yCd78N7TlVtPOHjeYv1QBEy39TjZ647rcHqLh0,345
 onnx_diagnostic/doc.py,sha256=t3RELgfooYnVMAi0JSpggWkQEgUsREz8NmRvn0TnLI8,2829
 onnx_diagnostic/ext_test_case.py,sha256=Bq4vdlM0P72H1orlKJTeOBqm1YGHTK-ylAlNsBe4LeA,43438
@@ -9,6 +9,7 @@ onnx_diagnostic/export/dynamic_shapes.py,sha256=HYf2OEi7PmRSj8uxMD-wbdVxxejkWXTP
 onnx_diagnostic/export/shape_helper.py,sha256=EQXHRVxwGpHRYhx8Y44Crqs640pmaIuKSwW1KJOW0IU,7501
 onnx_diagnostic/export/validate.py,sha256=_PGUql2DJhIgGKo0WjTGUc5AgsZUx8fEs00MePy-w98,6043
 onnx_diagnostic/helpers/__init__.py,sha256=GJ2GT7cgnlIveVUwMZhuvUwidbTJaKv8CsSIOpZDsJg,83
+onnx_diagnostic/helpers/_log_helper.py,sha256=OCxmJKBN3T4joxlptvfe18xNlyClQH6mrJ2OOQvxZtI,16686
 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=TeBUuGvqIMO-dsLDy7keaVt3ImZeifldwTgx6TEjBo8,11595
@@ -16,7 +17,7 @@ onnx_diagnostic/helpers/config_helper.py,sha256=9h1NWC9RLmu43Yf5Cz9usjMdLiyLWXMh
 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=_6K0IvfK7ymBE8uWFAOA1ksU_fMvl2BRtlxj5SA9R2I,58203
-onnx_diagnostic/helpers/log_helper.py,sha256=e89MI_i6PFBshm1cOOX5yCowEPIKzneMyCrpc34vpU0,77613
+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=p0Xh2Br38xAqUjB2214GiNOIbCgiVZKeyVEnjdyqyFI,21091
 onnx_diagnostic/helpers/model_builder_helper.py,sha256=RvDyPFqRboEU3HsQV_xi9oy-o3_4KuGFVzs5MhksduY,12552
@@ -73,7 +74,7 @@ onnx_diagnostic/reference/torch_ops/shape_ops.py,sha256=pJrNR2UB4PlWl6cv4EDl1uGl
 onnx_diagnostic/reference/torch_ops/unary_ops.py,sha256=E8Ys1eZsOTsucBKoXb1_Kl5LbBDygniDvW2BvN4IPMo,1708
 onnx_diagnostic/tasks/__init__.py,sha256=0BYtrAnr0zKN3om71oi-OVz5wFYDp9WWIk51qWjjyCw,2450
 onnx_diagnostic/tasks/automatic_speech_recognition.py,sha256=H94rxeiAjcJdECA1g95G_U9fZfpXk6dfjNKKYuvc4Qc,7130
-onnx_diagnostic/tasks/feature_extraction.py,sha256=MptOP-1ZSIMTIJ0woSqKLR7TlB9m7kww9V8cfEgZJTY,2502
+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=-vbZMA_ruo0WR_96YMYRvoNfq1plpElBJWXC2klAf7Q,7802
@@ -81,13 +82,13 @@ onnx_diagnostic/tasks/mixture_of_expert.py,sha256=DgIsbwzV4smysOK83wny91k3ix1Qt2
 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=Pk-H_qX5Y-2dzk45N9jbQ73S3O_d5-D11MyUhUfUwuM,8685
+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=tW9Gnum_eck3czNyctuUISA-Ek7pO37v5-11GC8QBW8,13124
 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=ZsUSOnKxeq4-dP86c5dTIbHMJFy_y690vvU4yfo6tNs,23438
+onnx_diagnostic/torch_export_patches/onnx_export_errors.py,sha256=NVJNhwr5LVwKzU9RsR2b8fOounmQ1W3nx_k-6XirASc,23701
 onnx_diagnostic/torch_export_patches/onnx_export_serialization.py,sha256=wFE2fNxihAA3iua79AEB97_RBVv4wvGxwS9g4RJaSIc,10715
 onnx_diagnostic/torch_export_patches/patch_expressions.py,sha256=vr4tt61cbDnaaaduzMj4UBZ8OUtr6GfDpIWwOYqjWzs,3213
 onnx_diagnostic/torch_export_patches/patch_inputs.py,sha256=9b4pmyT00BwLqi7WG-gliep1RUy3gXEgW6BDnlSSA-M,7689
@@ -97,26 +98,26 @@ onnx_diagnostic/torch_export_patches/eval/__init__.py,sha256=57x62uZNA80XiWgkG8F
 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=nVOg69e_cXvpcP5WIW9pIHCgnF-P_Ne87mRC6ep0g-I,45847
+onnx_diagnostic/torch_export_patches/patches/patch_transformers.py,sha256=ZRUHF_wamXDq6rn6V1ja5MqtZW_sJzgxdE5P900GHbI,48511
 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=9u2jkqnuyBkIF3R2sDEO0Jlkedl-cQhBNXxXXDLSEwE,8885
 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=dlWeRNLcQ2h3fxx07MA0NHoOPqT3-Afejgrj2Rozvck,63796
+onnx_diagnostic/torch_models/validate.py,sha256=FNqi-Dg7-S_dSQDdcUjc5ERgTquiXclwGv0UZxVHYbY,64866
 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_cached_configs.py,sha256=zZvIxTbmL55x44kCj3-T5Kg3Qzm9KB_Xj-MCcU9-LuQ,268245
-onnx_diagnostic/torch_models/hghub/model_inputs.py,sha256=qDw03KsLd_ZAMHBso--rUriCAZIewKFZG9n4-1zvGo8,10825
+onnx_diagnostic/torch_models/hghub/model_inputs.py,sha256=SDRLCA2zivEHIKr2RRRP-dZNiNUcpYS3EgP0unLExxY,11046
 onnx_diagnostic/torch_models/untrained/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 onnx_diagnostic/torch_models/untrained/llm_phi2.py,sha256=ynBTDHJHCk44NjLT_t6OiFDBdPP0rFGPteiONDxvztw,3708
 onnx_diagnostic/torch_models/untrained/llm_tiny_llm.py,sha256=QXw_Bs2SzfeiQMf-tmtVl83SmVOL4-Um7Qy-f0E48QI,2507
 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-0.7.4.dist-info/licenses/LICENSE.txt,sha256=Vv6TXglX6Rc0d-f8aREhayhT-6PMQXEyOmI2NKlUCMc,1045
-onnx_diagnostic-0.7.4.dist-info/METADATA,sha256=6l0XAH0UYEmqkPDswFSZFGrU5hYVEDB8YByQCiBkwlI,7431
-onnx_diagnostic-0.7.4.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-onnx_diagnostic-0.7.4.dist-info/top_level.txt,sha256=KwNkXewmcobM3ZT1DJLVWH6ebJzA5qKg7cWqKfpGNT4,16
-onnx_diagnostic-0.7.4.dist-info/RECORD,,
+onnx_diagnostic-0.7.5.dist-info/licenses/LICENSE.txt,sha256=Vv6TXglX6Rc0d-f8aREhayhT-6PMQXEyOmI2NKlUCMc,1045
+onnx_diagnostic-0.7.5.dist-info/METADATA,sha256=sTLrt4adKj6mufYnekwgSwSXxhgJzrV_TZTL-lHA0EU,7431
+onnx_diagnostic-0.7.5.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+onnx_diagnostic-0.7.5.dist-info/top_level.txt,sha256=KwNkXewmcobM3ZT1DJLVWH6ebJzA5qKg7cWqKfpGNT4,16
+onnx_diagnostic-0.7.5.dist-info/RECORD,,