up

Author: metascroy

.github/workflows/torchao_experimental_test.yml

@@ -35,9 +35,9 @@ jobs:
           conda activate venv
           pip install --extra-index-url "https://download.pytorch.org/whl/nightly/cpu" torch=="2.6.0.dev20250104"
           pip install numpy
+          pip install pytest
           USE_CPP=1 pip install .
       - name: Run tests
         run: |
           conda activate venv
-          python torchao/experimental/tests/test_packed_linear_int8_dynamic_activation_intx_weight_layout.py
-          python torchao/experimental/tests/test_q_dq_layout.py
+          pytest torchao/experimental/tests/test_int8_dynamic_activation_intx_weight.py

torchao/experimental/q_dq_layout.py

@@ -54,36 +54,6 @@ def _linear_impl(input_tensor, weight_tensor, bias):
         weight_tensor = weight_tensor.dequantize()
     return torch.nn.functional.linear(input_tensor, weight_tensor, bias)
 
-    # assert isinstance(input_tensor, )
-    # if isinstance(input_tensor, AffineQuantizedTensor):
-
-    #         input_tensor = input_tensor.dequantize()
-    #     if isinstance(weight_tensor, AffineQuantizedTensor):
-    #         weight_tensor = weight_tensor.dequantize()
-    #     return torch.nn.functional.linear(input_tensor, weight_tensor, bias)
-
-    # x = per_token_dynamic_quant(input_tensor)
-
-    # w_int_data = weight_tensor.tensor_impl.int_data
-    # w_scale = weight_tensor.tensor_impl.scale
-    # w_zero_point = weight_tensor.tensor_impl.zero_point
-    # assert len(weight_tensor.block_size) == 2
-    # assert weight_tensor.block_size[0] == 1
-    # group_size = weight_tensor.block_size[1]
-
-    # w_dq = _quantized_decomposed_dequantize_per_channel_group_wrapper(
-    #     w_int_data,
-    #     w_scale,
-    #     w_zero_point,
-    #     weight_tensor.quant_min,
-    #     weight_tensor.quant_max,
-    #     torch.int8,
-    #     group_size,
-    #     torch.float32,
-    # )
-
-    # return torch.nn.functional.linear(x, w_dq, bias)
-
 
 register_aqt_quantized_linear_dispatch(
     _linear_check,