[QNN] Use sigmoid Lookup Table method instead of fallback to fp32 (ap…

…ache#12038)
cjia4 · Jul 12, 2022 · ef5c3ed · ef5c3ed
1 parent 6d676ba
commit ef5c3ed
Show file tree

Hide file tree

Showing 2 changed files with 9 additions and 1 deletion.
diff --git a/python/tvm/relay/frontend/pytorch.py b/python/tvm/relay/frontend/pytorch.py
@@ -1567,7 +1567,7 @@ def func(x):
             assert len(inputs) == 3, "Input quant param not found in op inputs"
             input_scale = _expr.const(inputs[1])
             input_zero_point = _expr.const(inputs[2])
-            return qnn_torch.apply_with_fp32_fallback(data, input_scale, input_zero_point, func)
+            return qnn_torch.quantized_sigmoid(data, input_scale, input_zero_point)
 
         return func(data)
 

diff --git a/python/tvm/relay/frontend/qnn_torch.py b/python/tvm/relay/frontend/qnn_torch.py
@@ -571,6 +571,14 @@ def quantized_relu(data, input_zero_point):
     return _op.tensor.maximum(data, zp)
 
 
+def quantized_sigmoid(data, input_scale, input_zero_point):
+    output_scale = input_scale
+    output_zero_point = input_zero_point
+    return relay.qnn.op.sigmoid(
+        data, input_scale, input_zero_point, output_scale, output_zero_point
+    )
+
+
 def _quantize_per_tensor():
     def _impl(inputs, _):
         dim = len(infer_shape(inputs[0]))