diff options
Diffstat (limited to 'src/mlia/nn/tensorflow/optimizations/quantization.py')
| -rw-r--r-- | src/mlia/nn/tensorflow/optimizations/quantization.py | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/src/mlia/nn/tensorflow/optimizations/quantization.py b/src/mlia/nn/tensorflow/optimizations/quantization.py index 4e3c2c2..02c1d08 100644 --- a/src/mlia/nn/tensorflow/optimizations/quantization.py +++ b/src/mlia/nn/tensorflow/optimizations/quantization.py @@ -68,7 +68,7 @@ def quantize( ), f"Input tensor for quantization is of type {tensor.dtype}, but it must be float." quantized_tensor = (tensor / quant_params.scales) + quant_params.zero_points - quantized_tensor = np.clip( # type: ignore + quantized_tensor = np.clip( quantized_tensor, -128, 127, dtype=np.int8, casting="unsafe" ) return cast(np.ndarray, quantized_tensor) |