Rename Float8ActivationInt4WeightConfig to Float8DynamicActivationInt4WeightConfig (#2746)

Summary:
This was a mistake, we need to align the name with other dynamic quant configs

Test Plan:
CI

Reviewers:

Subscribers:

Tasks:

Tags:

Author: jerryzh168

docs/source/api_ref_quantization.rst

@@ -24,7 +24,7 @@ Inference APIs for quantize\_
     :nosignatures:
 
     Int4WeightOnlyConfig
-    Float8ActivationInt4WeightConfig
+    Float8DynamicActivationInt4WeightConfig
     Float8DynamicActivationFloat8WeightConfig
     Float8WeightOnlyConfig
     Float8StaticActivationFloat8WeightConfig

test/quantization/quantize_/workflows/int4/test_int4_preshuffled_tensor.py

@@ -16,7 +16,7 @@
 )
 
 from torchao.quantization import (
-    Float8ActivationInt4WeightConfig,
+    Float8DynamicActivationInt4WeightConfig,
     Int4WeightOnlyConfig,
     quantize_,
 )
@@ -33,7 +33,7 @@
     VERSION=2,
 )
 
-FP8_ACT_CONFIG = Float8ActivationInt4WeightConfig(
+FP8_ACT_CONFIG = Float8DynamicActivationInt4WeightConfig(
     group_size=128,
     packing_format="preshuffled",
 )

torchao/quantization/__init__.py

@@ -44,9 +44,9 @@
 from .quant_api import (
     CutlassInt4PackedLayout,
     FbgemmConfig,
-    Float8ActivationInt4WeightConfig,
     Float8DynamicActivationFloat8SemiSparseWeightConfig,
     Float8DynamicActivationFloat8WeightConfig,
+    Float8DynamicActivationInt4WeightConfig,
     Float8MMConfig,
     Float8StaticActivationFloat8WeightConfig,
     Float8WeightOnlyConfig,
@@ -143,7 +143,7 @@
     "Int8DynamicActivationInt8WeightConfig",
     "Int8DynamicActivationIntxWeightConfig",
     "Int4WeightOnlyConfig",
-    "Float8ActivationInt4WeightConfig",
+    "Float8DynamicActivationInt4WeightConfig",
     "Int8WeightOnlyConfig",
     "Float8WeightOnlyConfig",
     "Float8DynamicActivationFloat8WeightConfig",

torchao/quantization/quant_api.py

@@ -1252,7 +1252,7 @@ def _int4_weight_only_transform(
 
 
 @dataclass
-class Float8ActivationInt4WeightConfig(AOBaseConfig):
+class Float8DynamicActivationInt4WeightConfig(AOBaseConfig):
     """Configuration for apply float8 dynamic per row quantization and int4
     per group weight quantization to linear
 
@@ -1265,9 +1265,9 @@ class Float8ActivationInt4WeightConfig(AOBaseConfig):
     packing_format: PackingFormat = "preshuffled"
 
 
-@register_quantize_module_handler(Float8ActivationInt4WeightConfig)
-def _float8_activation_int4_weight_transform(
-    module: torch.nn.Module, config: Float8ActivationInt4WeightConfig
+@register_quantize_module_handler(Float8DynamicActivationInt4WeightConfig)
+def _float8_dynamic_activation_int4_weight_transform(
+    module: torch.nn.Module, config: Float8DynamicActivationInt4WeightConfig
 ) -> torch.nn.Module:
     assert hasattr(module, "weight"), (
         "applying int8 weight only quant requires module to have weight attribute"