use multi-turn batch=1 to export

Author: titaiwangms

.github/workflows/check-urls.yml

@@ -42,6 +42,6 @@ jobs:
         print_all: false
         timeout: 2
         retry_count# : 2
-        exclude_urls: https://hal.archives-,ouvertes.fr/hal-00990252/document,http://badge.fury.io/py/onnx-diagnostic,https://azure.microsoft.com/en-us/products/devops/pipelines,https://github.com/pytorch/pytorch/pull/83137#issuecomment-1211320670,https://github.com/NVIDIA/TransformerEngine.git@6a9edc38bf9b941b7d369af5103fa8fe0b121d61,https://medium.com/@msouza.os/llm-from-scratch-with-pytorch-9f21808c6319,https://github.com/pytorch/pytorch/blob/main/torch/fx/experimental/symbolic_shapes.py#L5965,https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-24-04.html,https://badge.fury.io/py/onnx-diagnostic.svg,https://github.com/huggingface/transformers/pull/36311
-        exclude_patterns: https://www.data.gouv.fr/fr/datasets/r/e3d83ab3-dc52-4c99-abaf-8a38050cc68c,https://dev.azure.com/,https://azure.microsoft.com/en-us/products/devops/pipelines,https://github.com/pytorch/pytorch/pull/83137#issuecomment-1211320670,https://github.com/NVIDIA/TransformerEngine.git@6a9edc38bf9b941b7d369af5103fa8fe0b121d61,https://github.com/pytorch/pytorch/blob/main/torch/,https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-24-04.html,https://badge.fury.io/py/onnx-diagnostic.svg,https://github.com/,https://codecov.io/,https://huggingface.co/
+        exclude_urls: https://hal.archives-,ouvertes.fr/hal-00990252/document,http://badge.fury.io/py/onnx-diagnostic,https://azure.microsoft.com/en-us/products/devops/pipelines,https://github.com/pytorch/pytorch/pull/83137#issuecomment-1211320670,https://github.com/NVIDIA/TransformerEngine.git@6a9edc38bf9b941b7d369af5103fa8fe0b121d61,https://medium.com/@msouza.os/llm-from-scratch-with-pytorch-9f21808c6319,https://github.com/pytorch/pytorch/blob/main/torch/fx/experimental/symbolic_shapes.py#L5965,https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-24-04.html,https://badge.fury.io/py/onnx-diagnostic.svg,https://github.com/huggingface/transformers/pull/36311,https://www.linux.org/
+        exclude_patterns: https://www.data.gouv.fr/fr/datasets/r/e3d83ab3-dc52-4c99-abaf-8a38050cc68c,https://dev.azure.com/,https://azure.microsoft.com/en-us/products/devops/pipelines,https://github.com/pytorch/pytorch/pull/83137#issuecomment-1211320670,https://github.com/NVIDIA/TransformerEngine.git@6a9edc38bf9b941b7d369af5103fa8fe0b121d61,https://github.com/pytorch/pytorch/blob/main/torch/,https://docs.nvidia.com/deeplearning/frameworks/pytorch-release-notes/rel-24-04.html,https://badge.fury.io/py/onnx-diagnostic.svg,https://github.com/,https://codecov.io/,https://huggingface.co/,https://www.linux.org/
         # force_pass : true

.github/workflows/ci.yml

@@ -17,7 +17,7 @@ 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', '4.56.1', 'main']
+        transformers: ['4.48.3', '4.51.3', '4.52.4', '4.55.4', '4.56.2', '4.57', 'main']
         torch: ['2.8', 'main']
         exclude:
           - python: '3.10'
@@ -26,30 +26,32 @@ jobs:
             transformers: 'main'
           - python: '3.10'
             transformers: '4.52.4'
-          - python: '3.10'
-            transformers: '4.53.3'
           - python: '3.10'
             transformers: '4.55.4'
           - python: '3.10'
-            transformers: '4.56.1'
+            transformers: '4.56.2'
+          - python: '3.10'
+            transformers: '4.57.0'
           - python: '3.11'
             torch: 'main'
-          - python: '3.11'
-            transformers: '4.53.3'
           - python: '3.11'
             transformers: 'main'
           - python: '3.11'
             transformers: '4.55.4'
           - python: '3.11'
-            transformers: '4.56.1'
+            transformers: '4.56.2'
+          - python: '3.11'
+            transformers: '4.57.0'
           - python: '3.13'
             torch: '2.8'
           - python: '3.13'
             transformers: '4.48.3'
           - python: '3.13'
             transformers: '4.51.3'
           - python: '3.13'
-            transformers: '4.52.4'
+            transformers: '4.55.4'
+          - python: '3.13'
+            transformers: '4.56.2'
     steps:
       - uses: actions/checkout@v3
 

CHANGELOGS.rst

@@ -1,9 +1,20 @@
 Change Logs
 ===========
 
+0.7.14
+++++++
+
+* :pr:`250`: add variables to track sequence nodes
+* :pr:`249`: patches _maybe_broadcast to support a corner case
+
 0.7.13
 ++++++
 
+* :pr:`247`: supports more gemma models with ModelBuilder
+* :pr:`246`: add a set of inputs checking models works for an empty cache on task text-generation
+* :pr:`237`: dummy inputs for google/gemma-3-4b-it
+* :pr:`244`: add a patch to bypass the exception raised when the dynamic dimension is in {0,1}
+
 0.7.12
 ++++++
 

_doc/conf.py

@@ -114,6 +114,8 @@ def linkcode_resolve(domain, info):
 nitpicky = True
 # See also scikit-learn/scikit-learn#26761
 nitpick_ignore = [
+    ("py:class", "_DimHint"),
+    ("py:class", "KeyPath"),
     ("py:class", "ast.Node"),
     ("py:class", "dtype"),
     ("py:class", "False"),

_doc/examples/plot_export_tiny_phi2.py

@@ -88,8 +88,9 @@
 # Shapes may not match on the second call with the modified inputs.
 
 
-with torch_export_patches(patch_transformers=True), torch.fx.experimental._config.patch(
-    backed_size_oblivious=True
+with (
+    torch_export_patches(patch_transformers=True),
+    torch.fx.experimental._config.patch(backed_size_oblivious=True),
 ):
 
     # Two unnecessary steps but useful in case of an error

_doc/index.rst

@@ -239,6 +239,7 @@ The function replaces dynamic dimensions defined as strings by
 Older versions
 ==============
 
+* `0.7.14 <../v0.7.14/index.html>`_
 * `0.7.13 <../v0.7.13/index.html>`_
 * `0.7.12 <../v0.7.12/index.html>`_
 * `0.7.11 <../v0.7.11/index.html>`_

_doc/patches.rst

@@ -91,7 +91,10 @@ Here is the list of available patches:
 
     for name, cls in p.__dict__.items():
         if name.startswith("patched_") and hasattr(cls, "_PATCHES_"):
-            print(f"{cls._PATCHED_CLASS_.__name__}: {', '.join(cls._PATCHES_)}")
+            print(
+                f"{cls._PATCHED_CLASS_.__name__}: "
+                f"{', '.join([_ for _ in cls._PATCHES_ if _ is not None])}"
+            )
 
 Cache serialization
 ===================

_doc/status/exported_program_dynamic.rst

@@ -53,6 +53,7 @@ with different options. This steps happens before converting into ONNX.
         for exporter in (
             "export-strict",
             "export-nostrict",
+            "export-nostrict-oblivious",
             "export-nostrict-decall",
             "export-tracing",
         ):

_doc/status/patches_coverage.rst

@@ -32,7 +32,10 @@ for transformers.
 
     for name, cls in p.__dict__.items():
         if name.startswith("patched_") and hasattr(cls, "_PATCHES_"):
-            print(f"{cls._PATCHED_CLASS_.__name__}: {', '.join(cls._PATCHES_)}")
+            print(
+                f"{cls._PATCHED_CLASS_.__name__}: "
+                f"{', '.join([_ for _ in cls._PATCHES_ if _ is not None])}"
+            )
 
 Half Automated Rewrites for Control Flows
 =========================================

_scripts/test_backend_onnxruntime.py

@@ -5,6 +5,7 @@
 import unittest
 import warnings
 from typing import Any
+import packaging.version as pv
 import numpy
 import onnx.backend.base
 import onnx.backend.test
@@ -140,6 +141,9 @@ def run_node(cls, node, inputs, device=None, outputs_info=None, **kwargs):
 
 backend_test.exclude("(test_adagrad|test_adam|test_add_uint8)")
 
+if pv.Version(onnxruntime.__version__) <= pv.Version("1.24"):
+    backend_test.exclude("(test_attention_4d_with|test_attention_4d_gqa)")
+
 
 # import all test cases at global scope to make them visible to python.unittest
 globals().update(backend_test.test_cases)