fix

Xilinx · Dec 14, 2024 · d3bb480 · d3bb480
1 parent 10fdfe1
commit d3bb480
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 5 deletions.
diff --git a/src/brevitas/export/inference/handler.py b/src/brevitas/export/inference/handler.py
@@ -66,7 +66,7 @@ def dequantize(self, x, scale, zero_point):
         return (x - zero_point) * scale
 
     def forward(self, x, unused_scale=None) -> Tuple[torch.Tensor]:
-        return self.dequantize(self.quantize(x)), self.scale, self.zero_point, self.bit_width
+        return self.dequantize(self.quantize(x, self.scale, self.zero_point), self.scale, self.zero_point), self.scale, self.zero_point, self.bit_width
 
 
 class IntWeightInferencetHandler(IntInferencetHandler):

diff --git a/src/brevitas/proxy/groupwise_int_runtime_quant.py b/src/brevitas/proxy/groupwise_int_runtime_quant.py
@@ -12,11 +12,10 @@ class GroupwiseActQuantProxyFromInjector(ActQuantProxyFromInjector):
     def __init__(self, quant_layer, quant_injector):
         super().__init__(quant_layer, quant_injector)
         self.cache_class = _CachedIOGroupwiseInt
-        self.group_dim = self.quant_injector.group_dim
 
-    # @property
-    # def group_dim(self):
-    #     return self.quant_injector.group_dim
+    @property
+    def group_dim(self):
+        return self.quant_injector.group_dim
 
     @property
     def group_size(self):