move to prototype folder

Author: jerryzh168

torchao/_models/llama/generate.py

@@ -227,7 +227,7 @@ def main(
         if "spinquant" in quantization:
             from torchao.prototype.spinquant import apply_spinquant
             apply_spinquant(model)
-        elif "int8wo" in quantization:
+        if "int8wo" in quantization:
             quantize_(model, int8_weight_only())
         elif "int8dq" in quantization:
             quantize_(model, int8_dynamic_activation_int8_weight())

torchao/quantization/autoquant_v2.py renamed to torchao/prototype/quantization/autoquant_v2.py

@@ -33,11 +33,11 @@
     benchmark_model,
 )
 
-from .granularity import (
+from torchao.quantization.granularity import (
     PerRow,
     PerTensor,
 )
-from .subclass import (  # noqa
+from torchao.quantization.subclass import (  # noqa
     Int8DynamicallyQuantizedLinearWeight,
     Int8WeightOnlyQuantizedLinearWeight,
     QuantizedLinearWeightBase,

torchao/quantization/subgraph_utils/extract_subgraphs.py renamed to torchao/prototype/quantization/subgraph_utils/extract_subgraphs.py

@@ -15,16 +15,10 @@
     OTHER_AUTOQUANT_CLASS_LIST,
     autoquant,
 )
-from .autoquant_v2 import (
+from torchao.prototype.quantization.autoquant_v2 import (
     DEFAULT_AUTOQUANT_CLASS_LIST as V2_DEFAULT_AUTOQUANT_CLASS_LIST,
-)
-from .autoquant_v2 import (
     DEFAULT_INT4_AUTOQUANT_CLASS_LIST as V2_DEFAULT_INT4_AUTOQUANT_CLASS_LIST,
-)
-from .autoquant_v2 import (
     OTHER_AUTOQUANT_CLASS_LIST as V2_OTHER_AUTOQUANT_CLASS_LIST,
-)
-from .autoquant_v2 import (
     autoquant_v2 as _autoquant_v2,
 )
 from .GPTQ import (

torchao/quantization/__init__.py

@@ -54,8 +54,8 @@
 )
 
 from .autoquant import AutoQuantizableLinearWeight, autoquant
-from .autoquant_v2 import AutoQuantizableLinearWeight as AutoQuantizableLinearWeightV2
-from .autoquant_v2 import autoquant_v2
+from torchao.prototype.quantization.autoquant_v2 import AutoQuantizableLinearWeight as AutoQuantizableLinearWeightV2
+from torchao.prototype.quantization.autoquant_v2 import autoquant_v2
 from .GPTQ import (
     Int4WeightOnlyGPTQQuantizer,
     Int4WeightOnlyQuantizer,