Fixed test cases fail

Signed-off-by: Jingyu Xin <[email protected]>

Author: jingyu-ml

modelopt/torch/quantization/nn/modules/tensor_quantizer.py

@@ -707,12 +707,12 @@ def _fake_quantize(self, inputs):
             outputs = scaled_e4m3(
                 inputs,
                 amax,
-                block_sizes_list,
                 self._get_bias(inputs),
                 E,
                 M,
                 self._trt_high_precision_dtype,
                 self._pass_through_bwd,
+                block_sizes_list,
             )
 
         else:

modelopt/torch/quantization/tensor_quant.py

@@ -417,12 +417,12 @@ def symbolic(
         g,
         inputs,
         amax=None,
-        block_sizes=None,
         bias=None,
         E=4,  # noqa: N803
         M=3,  # noqa: N803
         trt_high_precision_dtype=None,
         pass_through_bwd=False,
+        block_sizes=None,
     ):
         """ONNX symbolic function."""
         from .export_onnx import export_fp8
@@ -435,12 +435,12 @@ def forward(
         ctx,
         inputs,
         amax,
-        block_sizes,
         bias,
         E,  # noqa: N803
         M,  # noqa: N803
         trt_high_precision_dtype=None,
         pass_through_bwd=False,
+        block_sizes=None,
     ):
         """Forward method."""
         if E != 4 or M != 3: