More tests fix. No export

Xilinx · Jan 12, 2024 · 4d62dc9 · 4d62dc9
1 parent c383e5f
commit 4d62dc9
Show file tree

Hide file tree

Showing 4 changed files with 25 additions and 9 deletions.
diff --git a/src/brevitas/core/quant/int_base.py b/src/brevitas/core/quant/int_base.py
@@ -164,8 +164,8 @@ def forward(
             zero_point: Tensor,
             bit_width: Tensor,
             x: Tensor) -> Tensor:
-        y_int = self.to_int(pre_scale, pre_zero_point, bit_width, x)
-        y = y_int - zero_point
-        y = y * scale
+        y = self.to_int(pre_scale, pre_zero_point, bit_width, x)
+        # y = y_int - zero_point
+        # y = y * scale
         y = self.delay_wrapper(x, y)
         return y
diff --git a/src/brevitas/nn/mixin/parameter.py b/src/brevitas/nn/mixin/parameter.py
@@ -198,7 +198,11 @@ def quant_bias_zero_point(self):
         if self.bias is None:
             return None
         if not self.bias_quant.requires_input_scale and not self.bias_quant.requires_input_bit_width:
-            return self.bias_quant(self.bias).zero_point
+            bias_quant = self.bias_quant(self.bias)
+            if isinstance(bias_quant, QuantTensor):
+                return bias_quant.zero_point
+            else:
+                return None
         else:
             if self._cached_bias is None:
                 raise RuntimeError(

diff --git a/src/brevitas/nn/quant_rnn.py b/src/brevitas/nn/quant_rnn.py
@@ -418,11 +418,11 @@ def forward(self, inp, state):
         quant_input = self.maybe_quantize_input(inp)
         quant_weight_ih, quant_weight_hh, quant_bias = self.gate_params_fwd(
             self.gate_params, quant_input)
+        quant_input_value = _unpack_quant_tensor(quant_input)
         if getattr(quant_bias, 'value', quant_bias) is None:
-            quant_bias = torch.tensor(0., device=quant_input.value.device)
+            quant_bias = torch.tensor(0., device=quant_input_value.device)
         else:
             quant_bias = _unpack_quant_tensor(quant_bias)
-        quant_input_value = _unpack_quant_tensor(quant_input)
         quant_state = self.maybe_quantize_state(quant_input_value, state, self.cell.output_quant)
         if self.export_mode:
             cell = self.export_handler

diff --git a/tests/brevitas/fx/test_tracer.py b/tests/brevitas/fx/test_tracer.py
@@ -232,10 +232,22 @@ def test_module(module):
 @pytest.mark.parametrize('module', QUANT_TENSOR_MODULES)
 def test_quant_module(module):
     mod = module()
-    x = QuantTensor(torch.randn(INPUT_SIZE))
-    x_trace = QuantTensor(torch.randn(INPUT_SIZE))
+    x = QuantTensor(
+        torch.randint(-128, 127, INPUT_SIZE),
+        scale=0.1,
+        zero_point=0,
+        bit_width=8,
+        signed=True,
+        training=True)
+    x_trace = QuantTensor(
+        torch.randint(-128, 127, INPUT_SIZE),
+        scale=0.1,
+        zero_point=0,
+        bit_width=8,
+        signed=True,
+        training=True)
     with torch.no_grad():
         out = mod(x)
         graph_model = value_trace(mod, value_args={'x': x_trace})
         graph_out = graph_model(x)
-        assert graph_out.isclose(out).all().item()
+        assert torch.allclose(out, graph_out)