Per Row Int po2 kernel

Xilinx · Nov 26, 2024 · 5fa4e57 · 5fa4e57
1 parent 273338d
commit 5fa4e57
Show file tree

Hide file tree

Showing 2 changed files with 8 additions and 0 deletions.
diff --git a/src/brevitas_examples/common/generative/quantize.py b/src/brevitas_examples/common/generative/quantize.py
@@ -61,6 +61,7 @@
 from brevitas_examples.common.generative.quantizers import FP8e4m3OCPDynamicActPerRowFloat
 from brevitas_examples.common.generative.quantizers import Fp8e4m3WeightSymmetricGroupQuant
 from brevitas_examples.common.generative.quantizers import Int8DynamicActPerGroupFloat
+from brevitas_examples.common.generative.quantizers import Int8DynamicActPerRowFixedPoint
 from brevitas_examples.common.generative.quantizers import Int8DynamicActPerRowFloat
 from brevitas_examples.common.generative.quantizers import Int8DynamicActPerTensorFloat
 from brevitas_examples.common.generative.quantizers import IntWeightSymmetricGroupQuant
@@ -171,6 +172,8 @@
                         'sym': Int8DynamicActPerGroupFloat}}},
             'po2_scale': {
                 'stats': {
+                    'per_row': {
+                        'sym': Int8DynamicActPerRowFixedPoint,},
                     'per_group': {
                         'sym': MXInt8Act}}}}},
     'float': {

diff --git a/src/brevitas_examples/common/generative/quantizers.py b/src/brevitas_examples/common/generative/quantizers.py
@@ -81,6 +81,11 @@ class Int8DynamicActPerRowFloat(DynamicActProxyMixin, Int8ActPerTensorFloat):
     scaling_per_output_channel = True
 
 
+class Int8DynamicActPerRowFixedPoint(Int8DynamicActPerRowFloat):
+    restrict_scaling_type = RestrictValueType.POWER_OF_TWO
+    restrict_value_float_to_int_impl = FloorSte
+
+
 class Int8DynamicActPerGroupFloat(DynamicActProxyMixin, Int8ActPerTensorFloat):
     """
     Symmetric quantizer with per group scale.