Fix import error from quantization before PyTorch 2.3

Summary: pytorch#884 introduced a quantizer that was not available before
PyTorch 2.3, causing import errors for users using an earlier
version. In torchao, we gate the import by PyTorch version.
We can do the same here.

recipes/quantize.py

@@ -47,7 +47,7 @@ class QuantizationRecipe:
  multiple of groupsize.
  `percdamp`: GPTQ stablization hyperparameter, recommended to be .01
 
- 8da4w:
+ 8da4w (PyTorch 2.3+):
  torchtune.utils.quantization.Int8DynActInt4WeightQuantizer
  int8 per token dynamic activation with int4 weight only per axis group quantization
  Args:

torchtune/utils/quantization.py

@@ -11,15 +11,14 @@
  apply_weight_only_int8_quant,
  Int4WeightOnlyGPTQQuantizer,
  Int4WeightOnlyQuantizer,
- Int8DynActInt4WeightQuantizer,
  Quantizer,
 )
+from torchao.quantization.utils import TORCH_VERSION_AFTER_2_3
 
 __all__ = [
  "Int4WeightOnlyQuantizer",
  "Int4WeightOnlyGPTQQuantizer",
  "Int8WeightOnlyQuantizer",
- "Int8DynActInt4WeightQuantizer",
  "get_quantizer_mode",
 ]
 
@@ -36,10 +35,16 @@ def quantize(
  Int4WeightOnlyQuantizer: "4w",
  Int8WeightOnlyQuantizer: "8w",
  Int4WeightOnlyGPTQQuantizer: "4w-gptq",
- Int8DynActInt4WeightQuantizer: "8da4w",
 }
 
 
+if TORCH_VERSION_AFTER_2_3:
+ from torchao.quantization.quant_api import Int8DynActInt4WeightQuantizer
+
+ __all__.append("Int8DynActInt4WeightQuantizer")
+ _quantizer_to_mode[Int8DynActInt4WeightQuantizer] = "8da4w"
+
+
 def get_quantizer_mode(quantizer: Optional[Callable]) -> Optional[str]:
  """Given a quantizer object, returns a string that specifies the type of quantization e.g.
  4w, which means int4 weight only quantization.