Add support for groupwise quantization for int8 weight only quantization (#1121)

Summary:
This is to support deprecating torchchat int8 weight only quantization: https://github.com/pytorch/torchchat/blob/ecc628da7c32c486742d92a751ed045b2a2194be/torchchat/utils/quantize.py#L582

Test Plan:
python test/integration/test_integration.py -k test_weight_only_groupwise_quant

Reviewers:

Subscribers:

Tasks:

Tags:

Author: jerryzh168

test/integration/test_integration.py

@@ -109,6 +109,10 @@ def _int8wo_api(mod):
     else:
         change_linear_weights_to_int8_woqtensors(mod)
 
+def _int8wo_groupwise_api(mod):
+    group_size = 32
+    quantize_(mod, int8_weight_only(group_size=group_size), set_inductor_config=False)
+
 def _int8da_int8w_api(mod):
     if TORCH_VERSION_AT_LEAST_2_4:
         quantize_(mod, int8_dynamic_activation_int8_weight(), set_inductor_config=False)
@@ -927,6 +931,16 @@ def test_weight_only_quant(self):
             sqnr = compute_error(y_ref, y_wo)
             self.assertGreater(sqnr, 43.0)
 
+    def test_weight_only_groupwise_quant(self):
+        for x_shape in [[128, 512]]:
+            x = torch.randn(*x_shape)
+            m = nn.Sequential(nn.Linear(512, 32))
+            y_ref = m(x)
+            _int8wo_groupwise_api(m)
+            y_wo = m(x)
+            sqnr = compute_error(y_ref, y_wo)
+            self.assertGreater(sqnr, 45.0)
+
     @parameterized.expand(COMMON_DEVICE_DTYPE)
     @torch.no_grad()
     @unittest.skipIf(not torch.cuda.is_available(), "Need CUDA available")

torchao/quantization/quant_api.py

@@ -563,19 +563,22 @@ def apply_int4_weight_only_quant(weight):
     return _get_linear_subclass_inserter(apply_int4_weight_only_quant)
 
 
-def int8_weight_only():
+def int8_weight_only(group_size=None):
     """
     Applies int8 weight-only symmetric per-channel quantization to linear layers.
     """
-    def apply_int8wo_quant(weight):
+    def apply_int8wo_quant(weight, group_size=None):
         mapping_type = MappingType.SYMMETRIC
         target_dtype = torch.int8
         eps = torch.finfo(torch.float32).eps
         zero_point_dtype = torch.int64
+        if group_size is None:
+            group_size = weight.shape[1]
+
         block_size = (1, weight.shape[1])
         return to_affine_quantized_intx(weight, mapping_type, block_size, target_dtype, eps=eps, zero_point_dtype=zero_point_dtype)
 
-    return _get_linear_subclass_inserter(apply_int8wo_quant)
+    return _get_linear_subclass_inserter(apply_int8wo_quant, group_size=group_size)
 
 def _int8_symm_per_token_reduced_range_quant(x: torch.Tensor) -> torch.Tensor:
     mapping_type = MappingType.SYMMETRIC