diff --git a/docs/source/overview/core_concepts.rst b/docs/source/overview/core_concepts.rst
index 8c937c6..b18deeb 100644
--- a/docs/source/overview/core_concepts.rst
+++ b/docs/source/overview/core_concepts.rst
@@ -58,6 +58,28 @@ Use :func:`nsight.analyze.kernel` to annotate a benchmark function. Nsight Pytho
 
    benchmark(configs=[(1024,), (2048,)])
 
+Config values are mapped to function parameters in declaration order. The following
+parameter kinds are supported:
+
+- **Regular parameters** (``POSITIONAL_OR_KEYWORD``): fully supported.
+- **Keyword-only parameters** (after ``*`` or ``*args``): fully supported — config values
+  are automatically passed as keyword arguments.
+- **Parameters with default values**: configs may omit trailing parameters that have
+  defaults. The defaults are filled in automatically.
+- ``*args`` and ``**kwargs``: tolerated in the signature but ignored — they will always
+  be empty during profiling and will not appear in the results.
+
+.. code-block:: python
+
+   @nsight.analyze.kernel
+   def benchmark(x, y, *, mode="fast"):
+       ...
+
+   # mode has a default, so it can be omitted from configs.
+   # These two calls are equivalent:
+   benchmark(configs=[(1024, 2048)])
+   benchmark(configs=[(1024, 2048, "fast")])
+
 **3. Plot Decorator**  
 Add :func:`nsight.analyze.plot` to automatically generate plots from your profiling runs.
 
diff --git a/docs/source/release_notes/known_issues.rst b/docs/source/release_notes/known_issues.rst
index b97d061..0687c3b 100644
--- a/docs/source/release_notes/known_issues.rst
+++ b/docs/source/release_notes/known_issues.rst
@@ -43,3 +43,4 @@ Known Issues
 - Kernels launched from a subprocess which is created within the annotated region will not be profiled.
 - For the ``nsight.analyze.kernel``'s ``replay_mode="range"`` option, only a subset of CUDA APIs are supported within the annotated range. If an unsupported API call is detected, an error will be reported. For details on supported APIs, refer to the `NVIDIA Nsight Compute Profiling Guide <https://docs.nvidia.com/nsight-compute/ProfilingGuide/index.html#supported-apis>`_. In such cases, you can either switch to ``replay_mode="kernel"`` or modify the code to exclude the unsupported API from the annotated range.
 - Nested annotations (using ``nsight.annotate`` within another ``nsight.annotate`` context) are not supported. nsight-python errors out when nested annotations are used.
+- ``*args`` and ``**kwargs`` in decorated function signatures are tolerated but ignored — they will always be empty and will not appear in the profiling output.
diff --git a/nsight/collection/core.py b/nsight/collection/core.py
index 450d049..9e3997b 100644
--- a/nsight/collection/core.py
+++ b/nsight/collection/core.py
@@ -19,6 +19,87 @@
 from nsight import annotation, exceptions, thermovision, transformation, utils
 
 
+def _get_regular_params(
+    sig: inspect.Signature,
+) -> list[inspect.Parameter]:
+    """Return the list of regular (non-variadic) parameters from a signature."""
+    return [
+        p for p in sig.parameters.values()
+        if p.kind not in (inspect.Parameter.VAR_POSITIONAL, inspect.Parameter.VAR_KEYWORD)
+    ]
+
+
+def _count_params(
+    sig: inspect.Signature,
+) -> tuple[int, int]:
+    """Count (required, total) regular parameters in a signature.
+
+    Returns:
+        A (required_count, total_count) tuple. required_count excludes
+        parameters with defaults, total_count includes them.
+    """
+    params = _get_regular_params(sig)
+    required = sum(1 for p in params if p.default is inspect.Parameter.empty)
+    return required, len(params)
+
+
+def _pad_config_with_defaults(
+    sig: inspect.Signature, config: Sequence[Any]
+) -> tuple[Any, ...]:
+    """Pad a config tuple with default values for any missing trailing parameters.
+
+    If the config provides fewer values than the function has parameters, the
+    remaining parameters must have defaults, which are appended to the config.
+    This ensures the rest of the pipeline always sees full-length configs.
+
+    Args:
+        sig: The function's inspect.Signature.
+        config: The (possibly short) config tuple.
+
+    Returns:
+        A full-length config tuple with defaults filled in.
+    """
+    params = _get_regular_params(sig)
+    if len(config) == len(params):
+        return tuple(config)
+    padded = list(config)
+    for param in params[len(config):]:
+        padded.append(param.default)
+    return tuple(padded)
+
+
+def _bind_config_to_signature(
+    sig: inspect.Signature, config: Sequence[Any]
+) -> tuple[list[Any], dict[str, Any]]:
+    """Split a config tuple into positional args and keyword-only kwargs.
+
+    Maps config values to function parameters in declaration order, separating
+    them into positional arguments (for POSITIONAL_OR_KEYWORD params) and
+    keyword arguments (for KEYWORD_ONLY params). VAR_POSITIONAL and VAR_KEYWORD
+    parameters are skipped. Parameters with defaults that are not covered by
+    the config are left for the function to fill in.
+
+    Args:
+        sig: The function's inspect.Signature.
+        config: The config tuple to bind.
+
+    Returns:
+        A (positional_args, keyword_args) tuple ready for func(*pos, **kw).
+    """
+    positional: list[Any] = []
+    keyword: dict[str, Any] = {}
+    config_iter = iter(config)
+    for param in sig.parameters.values():
+        if param.kind in (inspect.Parameter.VAR_POSITIONAL, inspect.Parameter.VAR_KEYWORD):
+            continue
+        val = next(config_iter)
+        if param.kind == inspect.Parameter.KEYWORD_ONLY:
+            keyword[param.name] = val
+        else:
+            positional.append(val)
+    return positional, keyword
+
+
 def _sanitize_configs(
     func: Callable[..., Any],
     *args: Any,
@@ -60,6 +141,10 @@ def _sanitize_configs(
         - For functions with no parameters, an empty config [()] is created automatically.
         - The function combines `args` and `kwargs` into a single list if `args` are provided.
         - The function assumes that `kwargs` keys are in the expected order when combining.
+        - Config values are mapped to parameters in declaration order. Both
+          regular and keyword-only parameters are supported. Parameters with
+          defaults may be omitted from configs. ``*args`` and ``**kwargs``
+          are tolerated but ignored.
     """
     if len(args) > 0:
         # We do not expect any configs in this case
@@ -76,8 +161,8 @@ def _sanitize_configs(
         if decorator_configs is None:
             # Check if function takes no arguments
             sig = inspect.signature(func)
-            expected_arg_count = len(sig.parameters)
-            if expected_arg_count == 0:
+            required_count, total_count = _count_params(sig)
+            if required_count == 0:
                 # For functions with no arguments, create a single empty config
                 # This allows calling the function without requiring explicit configs
                 configs = [()]
@@ -105,8 +190,8 @@ def _sanitize_configs(
 
         # If function takes exactly one argument, allow scalar configs
         sig = inspect.signature(func)
-        expected_arg_count = len(sig.parameters)
-        if expected_arg_count == 1:
+        required_count, total_count = _count_params(sig)
+        if total_count == 1:
             normalized_configs: list[Sequence[Any]] = []
             for config in configs:
                 if utils.is_scalar(config):
@@ -122,11 +207,21 @@ def _sanitize_configs(
             )
         first_config_arg_count = config_lengths[0]
 
-        # Validate that the number of args matches the number of function parameters
-        if first_config_arg_count != expected_arg_count:
-            raise exceptions.ProfilerException(
-                f"Configs have {first_config_arg_count} arguments, but function expects {expected_arg_count}"
-            )
+        # Validate that the number of args is between required and total parameters
+        if not (required_count <= first_config_arg_count <= total_count):
+            if required_count == total_count:
+                raise exceptions.ProfilerException(
+                    f"Configs have {first_config_arg_count} arguments, but function expects {total_count}"
+                )
+            else:
+                raise exceptions.ProfilerException(
+                    f"Configs have {first_config_arg_count} arguments, but function expects "
+                    f"between {required_count} and {total_count}"
+                )
+
+        # Pad configs with default values for any missing trailing parameters
+        if first_config_arg_count < total_count:
+            configs = [_pad_config_with_defaults(sig, config) for config in configs]
 
     return configs  # type: ignore[return-value]
 
@@ -176,15 +271,16 @@ def run_profile_session(
     config_lengths: list[int] = list()
 
     for c in configs:
-        expected_arg_count = len(inspect.signature(func).parameters)
+        sig = inspect.signature(func)
+        required_count, total_count = _count_params(sig)
 
         # Handle scalar values
-        if expected_arg_count == 1:
+        if total_count == 1:
             if utils.is_scalar(c):
                 c = (c,)
 
         # Check if func supports the input configs
-        if expected_arg_count != len(c):
+        if not (required_count <= len(c) <= total_count):
             raise exceptions.ProfilerException(
                 f"Function '{func.__name__}' does not support the input configuration"
             )
@@ -211,8 +307,10 @@ def run_profile_session(
             # Clear active annotations before each run
             annotation.clear_active_annotations()
 
-            # Run the function with the config
-            result = func(*c)  # type: ignore[func-returns-value]
+            # Run the function with the config, splitting into positional
+            # and keyword-only args based on the function signature
+            pos_args, kw_args = _bind_config_to_signature(sig, c)
+            result = func(*pos_args, **kw_args)  # type: ignore[func-returns-value]
             if result is not None:
                 show_return_type_warning = True
 
diff --git a/nsight/extraction.py b/nsight/extraction.py
index f715301..2d8b3ab 100644
--- a/nsight/extraction.py
+++ b/nsight/extraction.py
@@ -123,8 +123,11 @@ def extract_df_from_report(
 
     sig = inspect.signature(func)
 
-    # Create a new array for each argument in the signature
-    arg_arrays: dict[str, list[Any]] = {name: [] for name in sig.parameters.keys()}
+    # Create a new array for each regular argument in the signature (exclude *args/**kwargs)
+    arg_arrays: dict[str, list[Any]] = {
+        name: [] for name, p in sig.parameters.items()
+        if p.kind not in (inspect.Parameter.VAR_POSITIONAL, inspect.Parameter.VAR_KEYWORD)
+    }
 
     # Extract all profiling data
     if output_progress:
@@ -255,9 +258,11 @@ def extract_df_from_report(
             all_metrics.append(tuple(metrics))
             hostnames.append(socket.gethostname())
             # Add a field for every config argument
-            bound_args = sig.bind(*conf)
-            for name, val in bound_args.arguments.items():
-                arg_arrays[name].append(val)
+            config_iter = iter(conf)
+            for name, param in sig.parameters.items():
+                if param.kind in (inspect.Parameter.VAR_POSITIONAL, inspect.Parameter.VAR_KEYWORD):
+                    continue
+                arg_arrays[name].append(next(config_iter))
 
     # Create the DataFrame with the initial columns
     df_data = {
diff --git a/nsight/transformation.py b/nsight/transformation.py
index 661d94a..a432ac2 100644
--- a/nsight/transformation.py
+++ b/nsight/transformation.py
@@ -37,8 +37,11 @@ def aggregate_data(
     if output_progress:
         print("[NSIGHT-PYTHON] Processing profiled data")
 
-    # Get the number of arguments in the signature of func
-    num_args = len(inspect.signature(func).parameters)
+    # Get the number of arguments in the signature of func (exclude *args/**kwargs)
+    num_args = sum(
+        1 for p in inspect.signature(func).parameters.values()
+        if p.kind not in (inspect.Parameter.VAR_POSITIONAL, inspect.Parameter.VAR_KEYWORD)
+    )
 
     # Get the last N fields of the dataframe where N is the number of arguments
     # Note: When num_args=0, we need an empty list (not all columns via [-0:])
@@ -104,8 +107,12 @@ def convert_non_sortable_columns(dframe: pd.DataFrame) -> pd.DataFrame:
                 )(col),
             )
 
+    # Ensure Value column is numeric (explode can leave it as object dtype
+    # even though the underlying values are always numeric from NCU metrics)
+    df["Value"] = pd.to_numeric(df["Value"])
+
     # Apply aggregation with named aggregation
-    groupby_df = df.groupby(groupby_columns + func_fields)
+    groupby_df = df.groupby(groupby_columns + func_fields, dropna=False)
     agg_df = groupby_df.agg(**named_aggs).reset_index()
 
     # Compute 95% confidence intervals
diff --git a/tests/test_profiler.py b/tests/test_profiler.py
index 9ec81e3..130e231 100644
--- a/tests/test_profiler.py
+++ b/tests/test_profiler.py
@@ -431,9 +431,13 @@ def function_with_default_parameter(x: int, y: Any = None) -> None:
 
 
 def test_function_with_default_parameter() -> None:
-    """Test that calling function with defaults without providing all args raises error."""
-    with pytest.raises(exceptions.ProfilerException):
-        function_with_default_parameter()
+    """Test that configs can omit parameters with defaults — they get padded automatically."""
+    result = function_with_default_parameter()
+    df = result.to_dataframe()
+
+    assert len(df) == 2, f"Expected 2 rows (2 configs), got {len(df)}"
+    # y has default None, which becomes NaN in pandas
+    assert df["y"].isna().all(), f"y should use default None/NaN, got {df['y'].tolist()}"
 
 
 # ============================================================================
@@ -1231,3 +1235,154 @@ def profiled_func(x: int, y: int) -> None:
         assert all(
             df["AvgValue"].notna() & (df["AvgValue"] > 0)
         ), f"Invalid AvgValue for metric {metrics}"
+
+
+# ============================================================================
+# Functions with **kwargs
+# ============================================================================
+
+
+def test_function_with_kwargs() -> None:
+    """Test that functions with **kwargs in their signature work correctly.
+
+    Regression test: _sanitize_configs counted **kwargs via len(sig.parameters),
+    causing a spurious validation error because the config arg count didn't
+    match the inflated parameter count.
+    """
+
+    @nsight.analyze.kernel(output="quiet")
+    def kernel_with_kwargs(x: int, y: int, **kwargs: Any) -> None:
+        a = torch.randn(x, y, device="cuda")
+        b = torch.randn(x, y, device="cuda")
+        with nsight.annotate("test_kwargs"):
+            _ = a + b
+
+    result = kernel_with_kwargs(configs=[(32, 32)])
+    df = result.to_dataframe()
+
+    assert len(df) == 1, f"Expected 1 row, got {len(df)}"
+    assert df["x"].iloc[0] == 32
+    assert df["y"].iloc[0] == 32
+
+
+def test_function_with_keyword_only_params() -> None:
+    """Test that functions with keyword-only parameters work correctly.
+
+    Keyword-only parameters (after * or *args) should be supported —
+    config values are mapped to parameters in declaration order and
+    passed as keyword arguments for keyword-only params.
+    """
+
+    @nsight.analyze.kernel(output="quiet")
+    def kernel_with_kw_only(x: int, *, y: int) -> None:
+        a = torch.randn(x, y, device="cuda")
+        b = torch.randn(x, y, device="cuda")
+        with nsight.annotate("test_kw_only"):
+            _ = a + b
+
+    result = kernel_with_kw_only(configs=[(32, 64)])
+    df = result.to_dataframe()
+
+    assert len(df) == 1, f"Expected 1 row, got {len(df)}"
+    assert df["x"].iloc[0] == 32
+    assert df["y"].iloc[0] == 64
+
+
+def test_function_with_args_and_keyword_only() -> None:
+    """Test that functions with *args and keyword-only params after it work."""
+
+    @nsight.analyze.kernel(output="quiet")
+    def kernel_mixed(x: int, *args: Any, y: int, **kwargs: Any) -> None:
+        a = torch.randn(x, y, device="cuda")
+        b = torch.randn(x, y, device="cuda")
+        with nsight.annotate("test_mixed"):
+            _ = a + b
+
+    result = kernel_mixed(configs=[(32, 64)])
+    df = result.to_dataframe()
+
+    assert len(df) == 1, f"Expected 1 row, got {len(df)}"
+    assert df["x"].iloc[0] == 32
+    assert df["y"].iloc[0] == 64
+
+
+def test_default_values_omitted() -> None:
+    """Test that configs can omit parameters that have default values."""
+
+    @nsight.analyze.kernel(output="quiet")
+    def kernel_with_defaults(x: int, y: int, z: int = 64) -> None:
+        a = torch.randn(x, y, device="cuda")
+        b = torch.randn(x, y, device="cuda")
+        with nsight.annotate("test_defaults"):
+            _ = a + b
+
+    result = kernel_with_defaults(configs=[(32, 32)])
+    df = result.to_dataframe()
+
+    assert len(df) == 1, f"Expected 1 row, got {len(df)}"
+    assert df["x"].iloc[0] == 32
+    assert df["y"].iloc[0] == 32
+    assert df["z"].iloc[0] == 64, f"z should use default 64, got {df['z'].iloc[0]}"
+
+
+def test_default_values_overridden() -> None:
+    """Test that configs can explicitly provide values for defaulted params."""
+
+    @nsight.analyze.kernel(output="quiet")
+    def kernel_with_defaults(x: int, y: int, z: int = 64) -> None:
+        a = torch.randn(x, y, device="cuda")
+        b = torch.randn(x, y, device="cuda")
+        with nsight.annotate("test_defaults_override"):
+            _ = a + b
+
+    result = kernel_with_defaults(configs=[(32, 32, 128)])
+    df = result.to_dataframe()
+
+    assert len(df) == 1, f"Expected 1 row, got {len(df)}"
+    assert df["z"].iloc[0] == 128, f"z should be overridden to 128, got {df['z'].iloc[0]}"
+
+
+def test_keyword_only_default_omitted() -> None:
+    """Test keyword-only params with defaults can be omitted from configs."""
+
+    @nsight.analyze.kernel(output="quiet")
+    def kernel_kw_default(x: int, *, y: int = 32) -> None:
+        a = torch.randn(x, y, device="cuda")
+        b = torch.randn(x, y, device="cuda")
+        with nsight.annotate("test_kw_default"):
+            _ = a + b
+
+    result = kernel_kw_default(configs=[(32,)])
+    df = result.to_dataframe()
+
+    assert len(df) == 1, f"Expected 1 row, got {len(df)}"
+    assert df["x"].iloc[0] == 32
+    assert df["y"].iloc[0] == 32, f"y should use default 32, got {df['y'].iloc[0]}"
+
+
+def test_too_few_config_args_rejected() -> None:
+    """Test that configs with fewer args than required params are rejected."""
+
+    @nsight.analyze.kernel(output="quiet")
+    def kernel_two_required(x: int, y: int, z: int = 64) -> None:
+        a = torch.randn(x, y, device="cuda")
+        b = torch.randn(x, y, device="cuda")
+        with nsight.annotate("test"):
+            _ = a + b
+
+    with pytest.raises(exceptions.ProfilerException, match="between 2 and 3"):
+        kernel_two_required(configs=[(32,)])
+
+
+def test_too_many_config_args_rejected() -> None:
+    """Test that configs with more args than total params are rejected."""
+
+    @nsight.analyze.kernel(output="quiet")
+    def kernel_two_params(x: int, y: int) -> None:
+        a = torch.randn(x, y, device="cuda")
+        b = torch.randn(x, y, device="cuda")
+        with nsight.annotate("test"):
+            _ = a + b
+
+    with pytest.raises(exceptions.ProfilerException, match="function expects 2"):
+        kernel_two_params(configs=[(32, 32, 99)])