adds a script to compare models (#221)

xadupre · web-flow · commit 15208c687593 · 2025-09-18T16:35:40.000+02:00
* script to compare models

* more stats

* manual seed

* moves slow import to a location where it works better

* lint
diff --git a/.github/workflows/documentation.yml b/.github/workflows/documentation.yml
@@ -118,9 +118,9 @@ jobs:
             grep ERROR doc.txt | grep -v 'l-plot-tiny-llm-export'
             exit 1
           fi
-          if [[ $(grep WARNING doc.txt | grep -v 'l-plot-tiny-llm-export' | grep -v 'Inline emphasis start-string' | grep -v 'Definition list ends without a blank line' | grep -v 'Unexpected section title or transition' | grep -v 'Inline strong start-string') ]]; then
+          if [[ $(grep WARNING doc.txt | grep -v 'l-plot-tiny-llm-export' | grep -v 'Inline emphasis start-string' | grep -v 'Definition list ends without a blank line' | grep -v 'Unexpected section title or transition' | grep -v 'Inline strong start-string' | grep -v 'MambaCache') ]]; then
             echo "Documentation produces warnings."
-            grep WARNING doc.txt | grep -v 'l-plot-tiny-llm-export' | grep -v 'Inline emphasis start-string' | grep -v 'Definition list ends without a blank line' | grep -v 'Unexpected section title or transition' | grep -v 'Inline strong start-string'
+            grep WARNING doc.txt | grep -v 'l-plot-tiny-llm-export' | grep -v 'Inline emphasis start-string' | grep -v 'Definition list ends without a blank line' | grep -v 'Unexpected section title or transition' | grep -v 'Inline strong start-string' | grep -v 'MambaCache'
             exit 1
           fi
 
diff --git a/_scripts/compare_model_execution.py b/_scripts/compare_model_execution.py
@@ -0,0 +1,131 @@
+"""
+Compares two ONNX models.
+"""
+
+print("-- import onnx")
+import onnx
+
+print("-- import onnx.helper")
+from onnx.helper import tensor_dtype_to_np_dtype
+
+print("-- import onnxruntime")
+import onnxruntime
+
+print("-- import torch")
+import torch
+
+print("-- import transformers")
+import transformers
+
+print("-- import huggingface_hub")
+import huggingface_hub
+
+print("-- import onnx-diagnostic.helper")
+from onnx_diagnostic.helpers.helper import flatten_object, string_type, max_diff, string_diff
+
+print("-- import onnx-diagnostic.torch_models.hghub")
+from onnx_diagnostic.torch_models.hghub import get_untrained_model_with_inputs
+
+print("-- done")
+
+model_id = "arnir0/Tiny-LLM"
+onnx1 = (
+    "dump_test/arnir0_Tiny-LLM-custom-default-f16-cuda-op20/"
+    "arnir0_Tiny-LLM-custom-default-f16-cuda-op20.onnx"
+)
+onnx2 = (
+    "dump_test/arnir0_Tiny-LLM-custom-default-f16-cuda-op21/"
+    "arnir0_Tiny-LLM-custom-default-f16-cuda-op21.onnx"
+)
+providers = ["CUDAExecutionProvider", "CPUExecutionProvider"]
+
+print(f"-- load {onnx1!r}")
+onx1 = onnx.load(onnx1)
+print(f"-- load {onnx2!r}")
+onx2 = onnx.load(onnx2)
+
+print(f"-- getting inputs for model_id {model_id!r}")
+data = get_untrained_model_with_inputs(model_id)
+inputs = data["inputs"]
+print(f"-- inputs: {string_type(inputs, with_shape=True)}")
+flatten_inputs = flatten_object(inputs, drop_keys=True)
+print(f"-- flat inputs: {string_type(flatten_inputs, with_shape=True)}")
+
+names = [i.name for i in onx1.graph.input]
+itypes = [i.type.tensor_type.elem_type for i in onx1.graph.input]
+assert names == [
+    i.name for i in onx2.graph.input
+], f"Not the same names for both models {names} != {[i.name for i in onx2.graph.input]}"
+feeds = {
+    n: t.numpy().astype(tensor_dtype_to_np_dtype(itype))
+    for n, itype, t in zip(names, itypes, flatten_inputs)
+}
+print(f"-- feeds: {string_type(feeds, with_shape=True)}")
+
+print(f"-- creating session 1 from {onnx1!r}")
+opts = onnxruntime.SessionOptions()
+opts.optimized_model_filepath = "debug1_full.onnx"
+opts.log_severity_level = 0
+opts.log_verbosity_level = 0
+sess1 = onnxruntime.InferenceSession(onnx1, opts, providers=providers)
+print(f"-- creating session 2 from {onnx2!r}")
+opts.optimized_model_filepath = "debug2_full.onnx"
+opts.log_severity_level = 0
+opts.log_verbosity_level = 0
+sess2 = onnxruntime.InferenceSession(onnx2, opts, providers=providers)
+
+print("-- run session1")
+expected1 = sess1.run(None, feeds)
+print(f"-- got {string_type(expected1, with_shape=True)}")
+print("-- run session2")
+expected2 = sess2.run(None, feeds)
+print(f"-- got {string_type(expected2, with_shape=True)}")
+
+print("-- compute differences")
+diff = max_diff(expected1, expected2)
+print(f"-- diff={string_diff(diff)}")
+
+
+def get_names(onx: onnx.ModelProto) -> list[str]:
+    names = []
+    for node in onx.graph.node:
+        for o in node.output:
+            names.append((o, node.op_type, node.name))
+    return names
+
+
+if diff["abs"] > 0.1:
+    print("--")
+    print("-- import select_model_inputs_outputs")
+    from onnx_extended.tools.onnx_nodes import select_model_inputs_outputs
+
+    print("-- looking into intermediate results")
+    names1 = get_names(onx1)
+    names2 = get_names(onx1)
+    common = [n for n in names1 if n in (set(names1) & set(names2))]
+    print(f"-- {len(common)} names / {len(names1)}-{len(names2)}")
+    print(f"-- first names {common[:5]}")
+    for name, op_type, op_name in common:
+        x1 = select_model_inputs_outputs(onx1, [name])
+        x2 = select_model_inputs_outputs(onx2, [name])
+        s1 = onnxruntime.InferenceSession(x1.SerializeToString(), providers=providers)
+        s2 = onnxruntime.InferenceSession(x2.SerializeToString(), providers=providers)
+        e1 = s1.run(None, feeds)
+        e2 = s2.run(None, feeds)
+        diff = max_diff(e1, e2)
+        print(
+            f"-- name={name!r}: diff={string_diff(diff)} "
+            f"- op_type={op_type!r}, op_name={op_name!r}"
+        )
+        if diff["abs"] > 0.1:
+            opts = onnxruntime.SessionOptions()
+            opts.optimized_model_filepath = "debug1.onnx"
+            onnxruntime.InferenceSession(x1.SerializeToString(), opts, providers=providers)
+            opts.optimized_model_filepath = "debug2.onnx"
+            onnxruntime.InferenceSession(x2.SerializeToString(), opts, providers=providers)
+            print("--")
+            print("-- break here")
+            print(f"-- feeds {string_type(feeds, with_shape=True)}")
+            print(f"-- e1={string_type(e1, with_shape=True, with_min_max=True)}")
+            print(f"-- e2={string_type(e2, with_shape=True, with_min_max=True)}")
+            break
diff --git a/onnx_diagnostic/_command_lines_parser.py b/onnx_diagnostic/_command_lines_parser.py
@@ -474,7 +474,7 @@ def get_parser_validate() -> ArgumentParser:
     )
     parser.add_argument(
         "--runtime",
-        choices=["onnxruntime", "torch", "ref"],
+        choices=["onnxruntime", "torch", "ref", "orteval", "orteval10"],
         default="onnxruntime",
         help="onnx runtime to use, `onnxruntime` by default",
     )
diff --git a/onnx_diagnostic/helpers/cache_helper.py b/onnx_diagnostic/helpers/cache_helper.py
@@ -4,11 +4,6 @@
 import transformers
 import transformers.cache_utils
 
-try:
-    from transformers.models.mamba.modeling_mamba import MambaCache
-except ImportError:
-    from transformers.cache_utils import MambaCache
-
 
 class CacheKeyValue:
     """
@@ -354,8 +349,15 @@ def make_encoder_decoder_cache(
     )
 
 
-def make_mamba_cache(key_value_pairs: List[Tuple[torch.Tensor, torch.Tensor]]) -> MambaCache:
+def make_mamba_cache(
+    key_value_pairs: List[Tuple[torch.Tensor, torch.Tensor]],
+) -> "MambaCache":  # noqa: F821
     "Creates a ``MambaCache``."
+    # import is moved here because this part is slow.
+    try:
+        from transformers.models.mamba.modeling_mamba import MambaCache
+    except ImportError:
+        from transformers.cache_utils import MambaCache
     dtype = key_value_pairs[0][0].dtype
 
     class _config:
diff --git a/onnx_diagnostic/torch_models/hghub/model_inputs.py b/onnx_diagnostic/torch_models/hghub/model_inputs.py
@@ -228,6 +228,8 @@ def get_untrained_model_with_inputs(
                     f"and use_pretrained=True."
                 )
 
+            seed = int(os.environ.get("SEED", "17"))
+            torch.manual_seed(seed)
             try:
                 if type(config) is dict:
                     model = cls_model(**config)
@@ -239,6 +241,8 @@ def get_untrained_model_with_inputs(
                 ) from e
 
     # input kwargs
+    seed = int(os.environ.get("SEED", "17")) + 1
+    torch.manual_seed(seed)
     kwargs, fct = random_input_kwargs(config, task)  # type: ignore[arg-type]
     if verbose:
         print(f"[get_untrained_model_with_inputs] use fct={fct}")
diff --git a/onnx_diagnostic/torch_models/validate.py b/onnx_diagnostic/torch_models/validate.py
@@ -7,8 +7,6 @@
 import time
 import numpy as np
 import onnx
-import onnxscript
-import onnxscript.rewriter.ort_fusions as ort_fusions
 import torch
 from ..export import CoupleInputsDynamicShapes
 from ..helpers import max_diff, string_type, string_diff
@@ -249,6 +247,7 @@ def _quiet_or_not_quiet(
         summary[f"time_{suffix}_latency_std"] = a.std()
         summary[f"time_{suffix}_latency_min"] = a.min()
         summary[f"time_{suffix}_latency_min"] = a.max()
+        summary[f"time_{suffix}_n"] = len(a)
     return res
 
 
@@ -337,7 +336,8 @@ def validate_model(
     :param subfolder: version or subfolders to uses when retrieving a model id
     :param opset: onnx opset to use for the conversion
     :param runtime: onnx runtime to use to check about discrepancies,
-        only if `do_run` is true
+        possible values ``onnxruntime``, ``torch``, ``orteval``,
+        ``orteval10``, ``ref`` only if `do_run` is true
     :param repeat: number of time to measure the model
     :param warmup: warmup the model first
     :param inputs2: checks that the second set of inputs is reunning as well,
@@ -364,7 +364,13 @@ def validate_model(
 
     The default runtime, :epkg:`onnxruntime` is used to validate a model and check the
     exported model returns the same outputs as the original one, otherwise,
-    :class:`onnx_diagnostic.reference.TorchOnnxEvaluator` is used.
+    :class:`onnx_diagnostic.reference.TorchOnnxEvaluator`
+    if ``runtime == 'torch'`` or
+    :class:`onnx_diagnostic.reference.OnnxruntimeEvaluator`
+    if ``runtime == 'orteval'`` or
+    :class:`onnx_diagnostic.reference.ExtendedReferenceEvaluator`
+    if ``runtime == 'ref'``,
+    ``orteval10`` increases the verbosity.
     """
     if isinstance(patch, bool):
         patch_kwargs = (
@@ -846,15 +852,24 @@ def node_iter(proto):
             raise NotImplementedError(f"Unexpected type={type(proto)}")
 
     counts: Dict[str, Union[float, int]] = {}
+    n_nodes = 0
+    n_nodes_nocst = 0
     for proto in node_iter(onx):
         if isinstance(proto, onnx.NodeProto):
             key = f"n_node_{proto.op_type}"
+            n_nodes += 1
+            if proto.op_type != "Constant":
+                n_nodes_nocst += 1
         else:
             key = f"n_node_initializer_{proto.data_type}"
 
         if key not in counts:
             counts[key] = 0
         counts[key] += 1
+
+    counts["n_node_nodes"] = n_nodes
+    counts["n_node_nodes_nocst"] = n_nodes_nocst
+    counts["n_node_functions"] = len(onx.functions)
     return counts
 
 
@@ -1155,7 +1170,7 @@ def validate_onnx_model(
     :param quiet: catch exception or not
     :param verbose: verbosity
     :param flavour: use a different version of the inputs
-    :param runtime: onnx runtime to use, onnxruntime or torch
+    :param runtime: onnx runtime to use, onnxruntime, torch, orteval, ref
     :param repeat: run that number of times the model
     :param warmup: warmup the model
     :param inputs2: to validate the model on the second input set
@@ -1202,23 +1217,66 @@ def _mk(key, flavour=flavour):
             f"{providers}..., flavour={flavour!r}"
         )
 
-    if runtime != "onnxruntime":
+    if runtime == "onnxruntime":
+        if os.environ.get("DUMPORTOPT", "") in ("1", "true", "True"):
+            opts = onnxruntime.SessionOptions()
+            opts.optimized_model_filepath = f"{data['onnx_filename']}.rtopt.onnx"
+            if verbose:
+                print(
+                    f"[validate_onnx_model] saved optimized onnxruntime "
+                    f"in {opts.optimized_model_filepath!r}"
+                )
+            onnxruntime.InferenceSession(data["onnx_filename"], opts, providers=providers)
+            if verbose:
+                print("[validate_onnx_model] -- done")
+
+        if verbose:
+            print("[validate_onnx_model] runtime is onnxruntime")
+        cls_runtime = lambda model, providers: onnxruntime.InferenceSession(
+            (model.SerializeToString() if isinstance(model, onnx.ModelProto) else model),
+            providers=providers,
+        )
+    elif runtime == "torch":
         from ..reference import TorchOnnxEvaluator
 
-    cls_runtime = (
-        (
-            lambda model, providers: onnxruntime.InferenceSession(
-                (model.SerializeToString() if isinstance(model, onnx.ModelProto) else model),
-                providers=providers,
+        if verbose:
+            print("[validate_onnx_model] runtime is TorchOnnxEvaluator")
+        cls_runtime = (
+            lambda model, providers, _cls_=TorchOnnxEvaluator: _cls_(  # type: ignore[misc]
+                model, providers=providers, verbose=max(verbose - 1, 0)
             )
         )
-        if runtime == "onnxruntime"
-        else (
-            lambda model, providers, _cls_=TorchOnnxEvaluator: _cls_(  # type: ignore[misc]
+    elif runtime == "orteval":
+        from ..reference import OnnxruntimeEvaluator
+
+        if verbose:
+            print("[validate_onnx_model] runtime is OnnxruntimeEvaluator")
+        cls_runtime = (
+            lambda model, providers, _cls_=OnnxruntimeEvaluator: _cls_(  # type: ignore[misc]
                 model, providers=providers, verbose=max(verbose - 1, 0)
             )
         )
-    )
+    elif runtime == "orteval10":
+        from ..reference import OnnxruntimeEvaluator
+
+        if verbose:
+            print("[validate_onnx_model] runtime is OnnxruntimeEvaluator(verbose=10)")
+        cls_runtime = (
+            lambda model, providers, _cls_=OnnxruntimeEvaluator: _cls_(  # type: ignore[misc]
+                model, providers=providers, verbose=10
+            )
+        )
+    elif runtime == "ref":
+        from ..reference import ExtendedReferenceEvaluator
+
+        if verbose:
+            print("[validate_onnx_model] runtime is ExtendedReferenceEvaluator")
+        cls_runtime = lambda model, providers, _cls_=ExtendedReferenceEvaluator: _cls_(  # type: ignore[misc]
+            model, verbose=max(verbose - 1, 0)
+        )
+    else:
+        raise ValueError(f"Unexpecteed runtime={runtime!r}")
+
     sess = _quiet_or_not_quiet(
         quiet,
         _mk("create_onnx_ort"),
@@ -1399,6 +1457,8 @@ def call_torch_export_onnx(
         if optimization == "ir":
             label, f_optim = "export_onnx_opt_ir", (lambda epo=epo: epo.optimize())
         else:
+            import onnxscript
+            import onnxscript.rewriter.ort_fusions as ort_fusions
 
             def _os_ort_optim(epo):
                 onnxscript.optimizer.optimize_ir(epo.model)
@@ -1683,6 +1743,9 @@ def call_torch_export_custom(
         print("[call_torch_export_custom] done (export)")
 
     if os_ort:
+        import onnxscript
+        import onnxscript.rewriter.ort_fusions as ort_fusions
+
         if verbose:
             print("[call_torch_export_custom] conversion to IR...")
         begin = time.perf_counter()
diff --git a/pyproject.toml b/pyproject.toml
@@ -28,6 +28,10 @@ disable_error_code = ["arg-type", "assignment", "import-untyped", "misc", "name-
 module = ["onnx_diagnostic.helpers.args_helper"]
 disable_error_code = ["arg-type", "call-overload", "index"]
 
+[[tool.mypy.overrides]]
+module = ["onnx_diagnostic.helpers.cache_helper"]
+disable_error_code = ["name-defined"]
+
 [[tool.mypy.overrides]]
 module = ["onnx_diagnostic.helpers.helper"]
 disable_error_code = ["arg-type", "assignment", "attr-defined", "call-overload", "misc", "name-defined", "union-attr"]
@@ -123,6 +127,7 @@ select = [
 "_doc/examples/plot_*.py" = ["E402", "B018", "PIE808", "SIM105", "SIM117"]
 "_doc/notebooks/plot_*.py" = ["E402", "B018", "PIE808", "SIM105", "SIM117"]
 "_doc/recipes/plot_*.py" = ["E402", "B018", "PIE808", "SIM105", "SIM117"]
+"_scripts/compare_model_execution.py" = ["E402", "F401"]
 "_doc/technical/plot_*.py" = ["E402", "B018", "PIE808", "SIM105", "SIM117"]
 "_unittests/*/test*.py" = ["B008", "B904", "PIE808", "SIM117", "SIM105", "UP008"]
 "onnx_diagnostic/export/__init__.py" = ["F401"]
@@ -131,6 +136,7 @@ select = [
 "onnx_diagnostic/reference/torch_ops/__init__.py" = ["F401"]
 "onnx_diagnostic/torch_models/hghub/__init__.py" = ["F401"]
 "onnx_diagnostic/torch_models/hghub/hub_data_cached_configs.py" = ["PIE804"]
+"onnx_diagnostic/torch_models/validate.py" = ["E731"]
 "onnx_diagnostic/torch_export_patches/__init__.py" = ["F401"]
 "onnx_diagnostic/torch_export_patches/patches/__init__.py" = ["F401"]
 "onnx_diagnostic/torch_models/llms.py" = ["F401"]

Original file line number	Diff line number	Diff line change
`@@ -474,7 +474,7 @@ def get_parser_validate() -> ArgumentParser:`
`474`	`474`	`)`
`475`	`475`	`parser.add_argument(`
`476`	`476`	`"--runtime",`
`477`		`- choices=["onnxruntime", "torch", "ref"],`
	`477`	`+ choices=["onnxruntime", "torch", "ref", "orteval", "orteval10"],`
`478`	`478`	`default="onnxruntime",`
`479`	`479`	help="onnx runtime to use, `onnxruntime` by default",
`480`	`480`	`)`