Add NPU (Ascend) backend support for INT4 weight-only quantization workflow

test/quantization/quantize_/workflows/int4/test_int4_plain_int32_tensor.py

@@ -35,7 +35,7 @@ def get_config(group_size):
 
 @unittest.skipIf(not torch_version_at_least("2.8.0"), "Need pytorch 2.8+")
 @unittest.skipIf(not torch.xpu.is_available(), "XPU not available")
-class Int4PlainInt32Tensor(TestCase):
+class Int4PlainInt32TensorXPU(TestCase):
     @parametrize(
         "sizes",
         [
@@ -98,8 +98,74 @@ def test_activation_prescaling(self):
         self.assertTrue(compute_error(original * _ACT_PRE_SCALE, quantized) > 20)
 
 
-instantiate_parametrized_tests(Int4PlainInt32Tensor)
+@unittest.skipIf(not torch_version_at_least("2.7.1"), "Need pytorch 2.7.1+")
+@unittest.skipIf(
+    torch.accelerator.current_accelerator().type != "npu"
+    or not torch.accelerator.is_available(),
+    "NPU not available",
+)
+class Int4PlainInt32TensorNPU(TestCase):
+    @parametrize("device", ["npu"])
+    @parametrize(
+        "sizes",
+        [
+            ((128,), 256, 128),
+            ((32, 128), 512, 128),
+            ((2, 32, 128), 256, 128),
+        ],
+    )
+    @parametrize("dtype", [torch.float16, torch.bfloat16])
+    @parametrize("group_size", [32, 64])
+    def test_linear(self, device, sizes, dtype, group_size):
+        M, N, K = sizes
+        input = torch.randn(*M, K, dtype=dtype, device=device)
+        linear = torch.nn.Linear(K, N, dtype=dtype, device=device)
+        orig_output = linear(input)
+        quantize_(linear, get_config(group_size))
+        quantized_output = linear(input)
+        self.assertTrue(compute_error(orig_output, quantized_output) > 10)
+
+    @parametrize("device", ["npu"])
+    @parametrize("dtype", [torch.float16, torch.bfloat16])
+    def test_module_path(self, device, dtype):
+        linear = torch.nn.Linear(128, 256, dtype=dtype, device=device)
+        quantize_(linear, get_config(group_size=64))
+        self.assertEqual(
+            str(type(linear.weight)),
+            "<class 'torchao.quantization.Int4PlainInt32Tensor'>",
+        )
+
+        with tempfile.NamedTemporaryFile() as f:
+            torch.save(linear.state_dict(), f)
+            f.seek(0)
+            state_dict = torch.load(f)
+            self.assertEqual(
+                str(type(state_dict["weight"])),
+                "<class 'torchao.quantization.Int4PlainInt32Tensor'>",
+            )
+
+    @parametrize("device", ["npu"])
+    @parametrize("dtype", [torch.float16, torch.bfloat16])
+    def test_activation_prescaling(self, device, dtype):
+        input = torch.randn(1, 128, dtype=dtype, device=device)
+        linear = torch.nn.Linear(128, 256, bias=False, dtype=dtype, device=device)
+        original = linear(input)
+        quantize_(linear, get_config(64))
+        qw = linear.weight
+        assert isinstance(qw, SupportsActivationPreScaling), (
+            "Expected int4 tensor supports activation prescaling"
+        )
+        assert qw.act_pre_scale is None, "Default `act_pre_scale` is None"
+        _ACT_PRE_SCALE = 2
+        qw.act_pre_scale = _ACT_PRE_SCALE
+        quantized = linear(input)
+
+        # making sure activation pre scaling is successfully applied to the activation
+        self.assertTrue(compute_error(original * _ACT_PRE_SCALE, quantized) > 10)
+
 
+instantiate_parametrized_tests(Int4PlainInt32TensorXPU)
+instantiate_parametrized_tests(Int4PlainInt32TensorNPU)
 
 if __name__ == "__main__":
     run_tests()

torchao/quantization/quantize_/workflows/int4/int4_plain_int32_tensor.py

@@ -14,9 +14,7 @@
     choose_qparams_affine,
     quantize_affine,
 )
-from torchao.utils import (
-    TorchAOBaseTensor,
-)
+from torchao.utils import TorchAOBaseTensor, torch_version_at_least
 
 __all__ = [
     "Int4PlainInt32Tensor",
@@ -91,58 +89,155 @@ def from_hp(
         w: torch.Tensor,
         block_size: List[int],
     ):
-        assert w.ndim == 2 and w.device.type == "xpu", (
-            f"Expecting 2D tensor on XPU, but got: {w.shape} on {w.device.type}"
-        )
-        assert len(block_size) == w.ndim
-        assert w.dtype in [torch.float16, torch.bfloat16], (
-            f"Expecting float16 or bfloat16 weight tensor, but got: {w.dtype}"
-        )
-        original_shape = w.shape
-        mapping_type = MappingType.ASYMMETRIC
-        target_dtype = torch.int32
-        quant_min = 0
-        quant_max = 15
-        eps = 1e-6
-        scale_dtype = None
-        zero_point_dtype = torch.int32
-        scale, zero_point = choose_qparams_affine(
-            w,
-            mapping_type,
-            block_size,
-            target_dtype,
-            quant_min,
-            quant_max,
-            eps,
-            scale_dtype,
-            zero_point_dtype,
-        )
-        int_data = quantize_affine(
-            w,
-            block_size,
-            scale,
-            zero_point,
-            target_dtype,
-            quant_min,
-            quant_max,
-        )
-        assert int_data.dtype == torch.int32, (
-            "torch.ops.aten._convert_weight_to_int4pack expects `int32` dtype"
-        )
-        packed_weight = (int_data[::, 1::2] << 4 | int_data[::, ::2]).to(torch.uint8)
-        packed_weight = torch.ops.aten._convert_weight_to_int4pack(
-            packed_weight.contiguous(), 8
-        )
-        scale = scale.reshape(int_data.shape[0], -1)
-        zero_point = zero_point.reshape(int_data.shape[0], -1)
-        return Int4PlainInt32Tensor(
-            packed_weight,
-            scale.transpose(0, 1).contiguous(),
-            zero_point.transpose(0, 1).contiguous().to(torch.int8),
-            block_size,
-            original_shape,
-            act_pre_scale=None,
-        )
+        if w.device.type == "xpu":
+            return _from_hp_xpu(cls, w, block_size)
+        elif w.device.type == "npu":
+            return _from_hp_npu(cls, w, block_size)
+        else:
+            raise NotImplementedError(
+                f"Int4PlainInt32Tensor does not support device '{w.device.type}' yet."
+            )
+
+
+def _from_hp_xpu(
+    cls,
+    w: torch.Tensor,
+    block_size: List[int],
+):
+    assert w.ndim == 2 and w.device.type == "xpu", (
+        f"Expecting 2D tensor on XPU, but got: {w.shape} on {w.device.type}"
+    )
+    assert len(block_size) == w.ndim
+    assert w.dtype in [torch.float16, torch.bfloat16], (
+        f"Expecting float16 or bfloat16 weight tensor, but got: {w.dtype}"
+    )
+    original_shape = w.shape
+    mapping_type = MappingType.ASYMMETRIC
+    target_dtype = torch.int32
+    quant_min = 0
+    quant_max = 15
+    eps = 1e-6
+    scale_dtype = None
+    zero_point_dtype = torch.int32
+    scale, zero_point = choose_qparams_affine(
+        w,
+        mapping_type,
+        block_size,
+        target_dtype,
+        quant_min,
+        quant_max,
+        eps,
+        scale_dtype,
+        zero_point_dtype,
+    )
+    int_data = quantize_affine(
+        w,
+        block_size,
+        scale,
+        zero_point,
+        target_dtype,
+        quant_min,
+        quant_max,
+    )
+    assert int_data.dtype == torch.int32, (
+        "torch.ops.aten._convert_weight_to_int4pack expects `int32` dtype"
+    )
+    packed_weight = (int_data[::, 1::2] << 4 | int_data[::, ::2]).to(torch.uint8)
+    packed_weight = torch.ops.aten._convert_weight_to_int4pack(
+        packed_weight.contiguous(), 8
+    )
+    scale = scale.reshape(int_data.shape[0], -1)
+    zero_point = zero_point.reshape(int_data.shape[0], -1)
+    return Int4PlainInt32Tensor(
+        packed_weight,
+        scale.transpose(0, 1).contiguous(),
+        zero_point.transpose(0, 1).contiguous().to(torch.int8),
+        block_size,
+        original_shape,
+        act_pre_scale=None,
+    )
+
+
+def _from_hp_npu(
+    cls,
+    w: torch.Tensor,
+    block_size: List[int],
+):
+    # Require PyTorch 2.7.1+ for NPU backend ops and backward compatibility.
+    assert torch_version_at_least("2.7.1"), (
+        "Need pytorch 2.7.1+ for NPU backend op support."
+    )
+
+    assert w.ndim == 2 and w.device.type == "npu", (
+        f"Expecting 2D tensor on NPU, but got: {w.shape} on {w.device.type}"
+    )
+    assert len(block_size) == w.ndim
+    assert w.dtype in [torch.float16, torch.bfloat16], (
+        f"Expecting float16 or bfloat16 weight tensor, but got: {w.dtype}"
+    )
+
+    group_size = block_size[1]
+    k_dim = w.shape[-1]
+    assert group_size >= 32 and group_size % 32 == 0 and group_size < k_dim, (
+        f"Invalid group_size={group_size}: "
+        f"expected to be a multiple of 32, "
+        f"in range [32, {k_dim - 1}] for per-group quantization, "
+        f"but got group_size={group_size} (k_dim={k_dim})."
+    )
+
+    original_shape = w.shape
+    mapping_type = MappingType.ASYMMETRIC
+    target_dtype = torch.int32
+    quant_min = -8
+    quant_max = 7
+    eps = 1e-6
+    scale_dtype = w.dtype
+    zero_point_dtype = w.dtype
+
+    scale, zero_point = choose_qparams_affine(
+        w,
+        mapping_type,
+        block_size,
+        target_dtype,
+        quant_min,
+        quant_max,
+        eps,
+        scale_dtype,
+        zero_point_dtype,
+    )
+
+    int_data = quantize_affine(
+        w,
+        block_size,
+        scale,
+        zero_point,
+        target_dtype,
+        quant_min,
+        quant_max,
+    )
+
+    assert int_data.dtype == torch.int32, (
+        "torch.ops.npu.npu_convert_weight_to_int4pack expects `int32` dtype"
+    )
+    assert int_data.shape[-1] % 8 == 0, (
+        f"torch.ops.npu.npu_convert_weight_to_int4pack expects last dim must be aligned to 8,but got {int_data.shape[-1]}"
+    )
+
+    packed_weight = torch.ops.npu.npu_convert_weight_to_int4pack(
+        int_data.contiguous(), 0
+    )
+
+    scale = scale.reshape(int_data.shape[0], -1)
+    zero_point = zero_point.reshape(int_data.shape[0], -1)
+
+    return Int4PlainInt32Tensor(
+        packed_weight.contiguous(),
+        scale.transpose(0, 1).contiguous(),
+        zero_point.transpose(0, 1).contiguous(),
+        block_size,
+        original_shape,
+        act_pre_scale=None,
+    )
 
 
 implements = Int4PlainInt32Tensor.implements
@@ -157,6 +252,22 @@ def _(func, types, args, kwargs):
         args[1],
         args[2] if len(args) > 2 else None,
     )
+
+    if input_tensor.device.type == "xpu":
+        return _linear_xpu(input_tensor, weight_tensor, bias)
+    elif input_tensor.device.type == "npu":
+        return _linear_npu(input_tensor, weight_tensor, bias)
+    else:
+        raise NotImplementedError(
+            f"Int4PlainInt32Tensor does not support device '{input_tensor.device.type}' yet."
+        )
+
+
+def _linear_xpu(
+    input_tensor,
+    weight_tensor,
+    bias,
+):
     assert input_tensor.device.type == "xpu", (
         f"For XPU device only but got: {input_tensor.device}"
     )
@@ -201,6 +312,71 @@ def _(func, types, args, kwargs):
     return y.to(orig_dtype)
 
 
+def _linear_npu(
+    input_tensor,
+    weight_tensor,
+    bias,
+):
+    assert input_tensor.device.type == "npu", (
+        f"For NPU device only but got: {input_tensor.device.type}"
+    )
+    assert isinstance(weight_tensor, Int4PlainInt32Tensor), (
+        f"Expected weight_tensor to be Int4PlainInt32NPUTensor, got: {type(weight_tensor)}"
+    )
+    assert weight_tensor.block_size[0] == 1, (
+        f"Requires groupwise quantization, got block_size: {weight_tensor.block_size}"
+    )
+    assert input_tensor.shape[-1] == weight_tensor.shape[1], (
+        f"Shapes of input and weight do not match, input:{input_tensor.shape}, weight: {weight_tensor.shape}"
+    )
+
+    if weight_tensor.act_pre_scale is not None:
+        input_tensor = input_tensor * weight_tensor.act_pre_scale
+
+    act_mat = input_tensor
+    packed_weight = weight_tensor.qdata
+    scale = weight_tensor.scale
+    zero_point = weight_tensor.zero_point
+
+    orig_act_size = act_mat.shape
+    orig_dtype = act_mat.dtype
+
+    # dtype alignment
+    if act_mat.dtype == torch.float16:
+        scale = scale.to(torch.float16)
+        zero_point = zero_point.to(torch.float16)
+        if bias is not None:
+            bias = bias.to(torch.float16)
+    elif act_mat.dtype == torch.bfloat16:
+        scale = scale.to(torch.bfloat16)
+        zero_point = zero_point.to(torch.bfloat16)
+        if bias is not None:
+            bias = bias.to(torch.float32)
+
+    # reshape to 2D
+    act_mat = act_mat.reshape(-1, act_mat.shape[-1])
+
+    # groupwise int4 quantization
+    groupsize = weight_tensor.block_size[1]
+
+    y = torch.ops.npu.npu_weight_quant_batchmatmul(
+        x=act_mat,
+        weight=packed_weight.transpose(-1, -2),
+        antiquant_scale=scale,
+        antiquant_offset=zero_point,
+        antiquant_group_size=groupsize,
+        bias=bias,
+    )
+
+    # remove out_feature padding
+    assert weight_tensor.ndim == 2
+    orig_out_features = weight_tensor.shape[-2]
+    y = y[:, :orig_out_features]
+    y = y.reshape(*orig_act_size[:-1], orig_out_features)
+
+    return y.to(orig_dtype)
+
+
 Int4PlainInt32Tensor.__module__ = "torchao.quantization"
 
 # Allow a model with Int4PlainInt32Tensor weights to be loaded with `weights_only=True`