Enable 6-bit kernel

Differential Revision: D63991820

Pull Request resolved: #1027

torchao/experimental/ops/linear_8bit_act_xbit_weight/op_linear_8bit_act_xbit_weight_aten.cpp

@@ -66,6 +66,7 @@ TORCH_LIBRARY(torchao, m) {
   DEFINE_OP(3);
   DEFINE_OP(4);
   DEFINE_OP(5);
+  DEFINE_OP(6);
 }
 
 TORCH_LIBRARY_IMPL(torchao, CPU, m) {
@@ -74,6 +75,7 @@ TORCH_LIBRARY_IMPL(torchao, CPU, m) {
   DEFINE_CPU_IMPL(3);
   DEFINE_CPU_IMPL(4);
   DEFINE_CPU_IMPL(5);
+  DEFINE_CPU_IMPL(6);
 }
 
 TORCH_LIBRARY_IMPL(torchao, Meta, m) {
@@ -82,4 +84,5 @@ TORCH_LIBRARY_IMPL(torchao, Meta, m) {
   DEFINE_META_IMPL(3);
   DEFINE_META_IMPL(4);
   DEFINE_META_IMPL(5);
+  DEFINE_META_IMPL(6);
 }

torchao/experimental/ops/linear_8bit_act_xbit_weight/op_linear_8bit_act_xbit_weight_executorch/w6s.cpp

@@ -0,0 +1,29 @@
+// Copyright (c) Meta Platforms, Inc. and affiliates.
+// All rights reserved.
+//
+// This source code is licensed under the license found in the
+// LICENSE file in the root directory of this source tree.
+
+// Unlike ATen, ExecuTorch op registration appears to only allow on
+// EXECUTORCH_LIBRARY per cpp file due to a name redefinition error, so a new
+// file is needed for each variant
+
+#include <torchao/experimental/ops/linear_8bit_act_xbit_weight/op_linear_8bit_act_xbit_weight-impl.h>
+
+namespace {
+Tensor _op_out(
+    RuntimeContext& ctx,
+    const Tensor& activations,
+    const Tensor& packed_weights,
+    const Tensor& group_size_tensor,
+    const Tensor& n_tensor,
+    const Tensor& k_tensor,
+    Tensor& out) {
+  (void)ctx;
+  linear_out_cpu</*weight_nbit*/ 6, /*has_weight_zeros*/ false>(
+      activations, packed_weights, group_size_tensor, n_tensor, k_tensor, out);
+  return out;
+}
+} // namespace
+
+EXECUTORCH_LIBRARY(torchao, "_linear_8bit_act_6bit0zp_weight.out", _op_out);

torchao/experimental/ops/linear_8bit_act_xbit_weight/op_linear_8bit_act_xbit_weight_executorch/w6sz.cpp

@@ -0,0 +1,29 @@
+// Copyright (c) Meta Platforms, Inc. and affiliates.
+// All rights reserved.
+//
+// This source code is licensed under the license found in the
+// LICENSE file in the root directory of this source tree.
+
+// Unlike ATen, ExecuTorch op registration appears to only allow on
+// EXECUTORCH_LIBRARY per cpp file due to a name redefinition error, so a new
+// file is needed for each variant
+
+#include <torchao/experimental/ops/linear_8bit_act_xbit_weight/op_linear_8bit_act_xbit_weight-impl.h>
+
+namespace {
+Tensor _op_out(
+    RuntimeContext& ctx,
+    const Tensor& activations,
+    const Tensor& packed_weights,
+    const Tensor& group_size_tensor,
+    const Tensor& n_tensor,
+    const Tensor& k_tensor,
+    Tensor& out) {
+  (void)ctx;
+  linear_out_cpu</*weight_nbit*/ 6, /*has_weight_zeros*/ true>(
+      activations, packed_weights, group_size_tensor, n_tensor, k_tensor, out);
+  return out;
+}
+} // namespace
+
+EXECUTORCH_LIBRARY(torchao, "_linear_8bit_act_6bit_weight.out", _op_out);

torchao/experimental/quant_api.py

@@ -115,15 +115,7 @@ def forward(self, x):
         lead_shape = x.shape[0:-2]
         m, k = x.shape[-2], x.shape[-1]
         n = self._n.shape[1]
-        x = x.reshape(-1, m, k)
-
-        res = [
-            self._linear_op(
-                x[i, :, :], self.packed_weights, self._group_size, self._n, self._k
-            )
-            for i in range(x.shape[0])
-        ]
-        res = torch.stack(res)
+        res = self._linear_op(x.reshape(-1, k), self.packed_weights, self._group_size, self._n, self._k)
         res = res.reshape(*lead_shape, m, n)
         return res
 
@@ -206,7 +198,7 @@ def forward(self, x):
 
 def _maybe_get_quantized_linear_native(nbit, has_weight_zeros):
     try:
-        if nbit in [1, 2, 3, 4, 5]:
+        if nbit in [1, 2, 3, 4, 5, 6]:
             wzp_suffix = "" if has_weight_zeros else "0zp"
             return _Int8DynActIntxWeightQuantizedLinearNative(
                 pack_weight_op=getattr(

torchao/experimental/tests/test_linear_8bit_act_xbit_weight_quantizer.py

@@ -36,7 +36,7 @@ def test_accuracy(self):
         m = 1
         n = 1071
         k = 4096
-        activations = torch.randn(m, k, dtype=torch.float32)
+        activations = torch.randn(2, 3, m, k, dtype=torch.float32)
         model = torch.nn.Sequential(*[torch.nn.Linear(k, n, bias=False)])
 
         for nbit in [1, 2, 3, 4, 5, 6, 7]:
@@ -84,7 +84,7 @@ def test_export_compile_aoti(self):
         layers = [torch.nn.Linear(k0, k1, bias=False), torch.nn.Linear(k1, k2, bias=False), torch.nn.Linear(k2, k3, bias=False)]
         model = torch.nn.Sequential(*layers)
 
-        activations = torch.randn(2, 1, m, k0, dtype=torch.float32)
+        activations = torch.randn(m, k0, dtype=torch.float32)
 
         print("Quantizing model")
         quantizer = Int8DynActIntxWeightQuantizer(