Fix Int4WeightEmbeddingQATQuantizer.convert path

**Summary:** Fixes the issue where
`Int4WeightEmbeddingQATQuantizer`'s convert path assigned the
scales and zero points to the wrong attributes ("scales" and
"zeros" instead of "scale" and "zero point"), and also ensures
the precisions are correctly set.

**Test Plan:**
python test/quantization/test_qat.py -k test_qat_4w_embedding

Author: andrewor14

test/quantization/test_qat.py

@@ -789,16 +789,40 @@ def test_composable_qat_quantizer(self):
     )
     def test_qat_4w_embedding(self):
         from torchao.quantization.qat import Int4WeightOnlyEmbeddingQATQuantizer
+        from torchao._executorch_ops import (
+            _quantized_decomposed_quantize_per_channel_group_wrapper,
+        )
 
+        group_size = 256
         model = M2()
         x = model.example_inputs()
         model(*x)
-        quantizer = Int4WeightOnlyEmbeddingQATQuantizer()
+        quantizer = Int4WeightOnlyEmbeddingQATQuantizer(group_size)
         prepared = quantizer.prepare(model)
+        prepared_embedding_weight = copy.deepcopy(prepared.embedding.weight)
         prepared(*x)
         converted = quantizer.convert(model)
         converted(*x)
 
+        # Assert the scales, zero points, and weights are correct after convert
+        qmin, qmax = -8, 7
+        (s, zp) = get_group_qparams_symmetric(
+            prepared_embedding_weight, 4, group_size,
+        )
+        zp = zp.to(torch.int32)
+        q_weight = _quantized_decomposed_quantize_per_channel_group_wrapper(
+            prepared_embedding_weight,
+            s,
+            zp,
+            qmin,
+            qmax,
+            torch.int8,
+            group_size,
+        )
+        torch.testing.assert_close(converted.embedding.weight, q_weight)
+        torch.testing.assert_close(converted.embedding.scale, s)
+        torch.testing.assert_close(converted.embedding.zero_point, zp)
+
     def test_fake_quantize_config_granularity(self):
         """
         Test initialization and property setting of `FakeQuantizeConfig`'s granularity.

torchao/quantization/qat/embedding.py

@@ -245,8 +245,8 @@ def _convert_helper(self, module: torch.nn.Module):
                     group_size,
                 )
                 quantized_embedding.weight = q_weight
-                quantized_embedding.scales = s
-                quantized_embedding.zeros = zp
+                quantized_embedding.scale = s.to(scale_precision)
+                quantized_embedding.zero_point = zp.to(zero_point_precision)
             else:
                 self._convert_helper(child)