fix: use float base in pow to avoid Inductor dtype mismatch (#627)

* fix: use float base in pow to avoid Inductor dtype mismatch

Signed-off-by: Bias92 <gongnobi@gmail.com>

* style: format after float base change

Signed-off-by: Bias92 <gongnobi@gmail.com>

---------

Signed-off-by: Bias92 <gongnobi@gmail.com>
Co-authored-by: Bias92 <gongnobi@gmail.com>
Co-authored-by: HDCharles <39544797+HDCharles@users.noreply.github.com>

Author: 3 people

src/compressed_tensors/quantization/utils/helpers.py

@@ -205,7 +205,7 @@ def calculate_range(
     :return: tuple endpoints for the given quantization range
     """
     if quantization_args.type == QuantizationType.INT:
-        bit_range = 2**quantization_args.num_bits
+        bit_range = 2.0**quantization_args.num_bits
         q_max = torch.tensor(bit_range / 2 - 1, device=device)
         q_min = torch.tensor(-bit_range / 2, device=device)
     elif quantization_args.type == QuantizationType.FLOAT: