litert-quantizer-nightly 0.10.0.dev20261008__py3-none-any.whl

litert_quantizer/__init__.py

@@ -0,0 +1,19 @@
+# Copyright 2024 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+
+"""Init file for the AI Edge quantizer package."""
+
+# pylint: disable=unused-import
+from .quantizer import Quantizer

litert_quantizer/algorithm_manager.py

@@ -0,0 +1,480 @@
+# Copyright 2024 Google LLC
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+# ==============================================================================
+
+"""Quantizer Algorithm Manager Interface."""
+
+import enum
+import functools
+import immutabledict
+from litert_quantizer import algorithm_manager_api
+from litert_quantizer import default_policy
+from litert_quantizer import qtyping
+from litert_quantizer.algorithms.nonlinear_quantize import float_casting
+from litert_quantizer.algorithms.uniform_quantize import common_quantize
+from litert_quantizer.algorithms.uniform_quantize import dequantized_weight_recovery
+from litert_quantizer.algorithms.uniform_quantize import gptq
+from litert_quantizer.algorithms.uniform_quantize import hadamard_rotation
+from litert_quantizer.algorithms.uniform_quantize import mse
+from litert_quantizer.algorithms.uniform_quantize import naive_min_max_quantize
+from litert_quantizer.algorithms.uniform_quantize import octav
+from litert_quantizer.algorithms.uniform_quantize import oscar
+from litert_quantizer.utils import qsv_utils
+
+
+immutabledict = immutabledict.immutabledict
+
+
+# TODO: b/399775701 - Clean up this file.
+
+_TFLOpName = qtyping.TFLOperationName
+
+_alg_manager_instance = algorithm_manager_api.AlgorithmManagerApi()
+
+# Expose instance functions.
+get_quantization_func = _alg_manager_instance.get_quantization_func
+get_supported_ops = _alg_manager_instance.get_supported_ops
+get_update_qsv_func = _alg_manager_instance.get_update_qsv_func
+get_init_qsv_func = _alg_manager_instance.get_init_qsv_func
+register_op_quant_config_validation_func = (
+    _alg_manager_instance.register_op_quant_config_validation_func
+)
+register_config_check_policy_func = (
+    _alg_manager_instance.register_config_check_policy
+)
+register_quantized_op = _alg_manager_instance.register_quantized_op
+is_op_registered = _alg_manager_instance.is_op_registered
+is_algorithm_registered = _alg_manager_instance.is_algorithm_registered
+check_op_quantization_config = (
+    _alg_manager_instance.check_op_quantization_config
+)
+
+
+# Quantization algorithms.
+class AlgorithmName(str, enum.Enum):
+  NO_QUANTIZE = "no_quantize"
+  MIN_MAX_UNIFORM_QUANT = naive_min_max_quantize.ALGORITHM_KEY
+  FLOAT_CASTING = float_casting.ALGORITHM_KEY
+  DEQUANTIZED_WEIGHT_RECOVERY = dequantized_weight_recovery.ALGORITHM_KEY
+  OCTAV = octav.ALGORITHM_KEY
+  HADAMARD_ROTATION = hadamard_rotation.CUSTOM_OP_ALGORITHM_KEY
+  DECOMPOSED_HADAMARD_ROTATION = hadamard_rotation.DECOMPOSED_ALGORITHM_KEY
+  MSE = mse.ALGORITHM_KEY
+  GPTQ = gptq.ALGORITHM_KEY
+  OSCAR = oscar.ALGORITHM_KEY
+
+
+### MIN/MAX_UNIFORM_QUANT ###
+
+# Register MIN_MAX_UNIFORM_QUANT algorithm.
+register_op_quant_config_validation_func(
+    AlgorithmName.MIN_MAX_UNIFORM_QUANT,
+    common_quantize.check_op_quantization_config,
+)
+
+# Register a config check policy for MIN_MAX_UNIFORM_QUANT algorithm.
+register_config_check_policy_func(
+    AlgorithmName.MIN_MAX_UNIFORM_QUANT,
+    default_policy.DEFAULT_CONFIG_CHECK_POLICY,
+)
+
+MIN_MAX_OP_NAME_MATERIALIZE_FUNC_DICT = {
+    _TFLOpName.INPUT: common_quantize.materialize_input,
+    _TFLOpName.OUTPUT: common_quantize.materialize_output,
+    _TFLOpName.FULLY_CONNECTED: common_quantize.materialize_fc_conv,
+    _TFLOpName.BATCH_MATMUL: common_quantize.materialize_batch_matmul,
+    _TFLOpName.CONV_2D: common_quantize.materialize_fc_conv,
+    _TFLOpName.DEPTHWISE_CONV_2D: common_quantize.materialize_fc_conv,
+    _TFLOpName.CONV_2D_TRANSPOSE: common_quantize.materialize_conv2d_transpose,
+    _TFLOpName.RESHAPE: common_quantize.materialize_reshape,
+    _TFLOpName.AVERAGE_POOL_2D: common_quantize.materialize_average_pool_2d,
+    _TFLOpName.EMBEDDING_LOOKUP: common_quantize.materialize_embedding_lookup,
+    _TFLOpName.SOFTMAX: common_quantize.materialize_softmax_and_logistic,
+    _TFLOpName.TANH: common_quantize.materialize_tanh,
+    _TFLOpName.TRANSPOSE: common_quantize.materialize_transpose,
+    _TFLOpName.GELU: common_quantize.materialize_gelu,
+    _TFLOpName.ADD: common_quantize.materialize_add,
+    _TFLOpName.SUB: common_quantize.materialize_sub,
+    _TFLOpName.MUL: common_quantize.materialize_mul,
+    _TFLOpName.MEAN: common_quantize.materialize_mean,
+    _TFLOpName.RSQRT: common_quantize.materialize_rsqrt,
+    _TFLOpName.CONCATENATION: common_quantize.materialize_concatenation,
+    _TFLOpName.STRIDED_SLICE: common_quantize.materialize_strided_slice,
+    _TFLOpName.SPLIT: common_quantize.materialize_split,
+    _TFLOpName.LOGISTIC: common_quantize.materialize_softmax_and_logistic,
+    _TFLOpName.SLICE: common_quantize.materialize_slice,
+    _TFLOpName.SUM: common_quantize.materialize_sum,
+    _TFLOpName.SELECT: common_quantize.materialize_select,
+    _TFLOpName.SELECT_V2: common_quantize.materialize_select_v2,
+    _TFLOpName.DYNAMIC_UPDATE_SLICE: (
+        common_quantize.materialize_dynamic_update_slice
+    ),
+    _TFLOpName.STABLEHLO_COMPOSITE: common_quantize.materialize_composite,
+    _TFLOpName.PAD: common_quantize.materialize_pad,
+    _TFLOpName.SQUARED_DIFFERENCE: (
+        common_quantize.materialize_squared_difference
+    ),
+    _TFLOpName.MAX_POOL_2D: common_quantize.materialize_max_pool_2d,
+    _TFLOpName.RESIZE_BILINEAR: common_quantize.materialize_resize_bilinear,
+    _TFLOpName.RESIZE_NEAREST_NEIGHBOR: (
+        common_quantize.materialize_resize_nearest_neighbor
+    ),
+    _TFLOpName.GATHER_ND: common_quantize.materialize_gather_nd,
+    _TFLOpName.PACK: common_quantize.materialize_pack,
+    _TFLOpName.UNPACK: common_quantize.materialize_unpack,
+    _TFLOpName.DIV: common_quantize.materialize_div,
+    _TFLOpName.BROADCAST_TO: common_quantize.materialize_broadcast_to,
+    _TFLOpName.SQRT: common_quantize.materialize_sqrt,
+    _TFLOpName.GATHER: common_quantize.materialize_gather,
+    _TFLOpName.HARD_SWISH: common_quantize.materialize_hard_swish,
+    _TFLOpName.MAXIMUM: common_quantize.materialize_maximum,
+    _TFLOpName.PADV2: common_quantize.materialize_padv2,
+    _TFLOpName.REDUCE_MIN: common_quantize.materialize_reduce_min,
+    _TFLOpName.EQUAL: common_quantize.materialize_equal,
+    _TFLOpName.NOT_EQUAL: common_quantize.materialize_not_equal,
+    _TFLOpName.MIRROR_PAD: common_quantize.materialize_mirror_pad,
+    _TFLOpName.SPACE_TO_DEPTH: common_quantize.materialize_space_to_depth,
+    _TFLOpName.RELU: common_quantize.materialize_relu,
+}
+for op_name, materialize_func in MIN_MAX_OP_NAME_MATERIALIZE_FUNC_DICT.items():
+  register_quantized_op(
+      AlgorithmName.MIN_MAX_UNIFORM_QUANT,
+      op_name,
+      naive_min_max_quantize.init_qsvs,
+      calibration_func=naive_min_max_quantize.min_max_calibrate,
+      # Most of the materialize op functions are common for all algorithms
+      # except for the function to get scale and zero point, i.e.,
+      # get_tensor_quant_params. So we use functools.partial here to pass in the
+      # common utility function and thealgorithm-specific function.
+      materialize_func=functools.partial(
+          materialize_func,
+          naive_min_max_quantize.get_tensor_quant_params,
+      ),
+  )
+
+### FLOAT_CASTING ###
+register_op_quant_config_validation_func(
+    AlgorithmName.FLOAT_CASTING,
+    float_casting.check_op_quantization_config,
+)
+
+# Register a config check policy for FLOAT_CASTING algorithm.
+# TODO: b/353780772 - Replace an empty policy for FLOAT_CASTING algorithm.
+register_config_check_policy_func(
+    AlgorithmName.FLOAT_CASTING, qtyping.ConfigCheckPolicyDict()
+)
+
+for op_name, materialize_func in zip(
+    (
+        _TFLOpName.FULLY_CONNECTED,
+        _TFLOpName.CONV_2D,
+        _TFLOpName.DEPTHWISE_CONV_2D,
+        _TFLOpName.CONV_2D_TRANSPOSE,
+        _TFLOpName.EMBEDDING_LOOKUP,
+    ),
+    (
+        float_casting.materialize_fc_conv,
+        float_casting.materialize_fc_conv,
+        float_casting.materialize_fc_conv,
+        float_casting.materialize_conv2d_transpose,
+        float_casting.materialize_embedding_lookup,
+    ),
+):
+  register_quantized_op(
+      AlgorithmName.FLOAT_CASTING,
+      op_name,
+      float_casting.init_qsvs,  # pyrefly: ignore[bad-argument-type]
+      calibration_func=float_casting.calibrate,  # pyrefly: ignore[bad-argument-type]
+      materialize_func=materialize_func,  # pyrefly: ignore[bad-argument-type]
+  )
+
+### DEQUANTIZED_WEIGHT_RECOVERY ###
+register_op_quant_config_validation_func(
+    AlgorithmName.DEQUANTIZED_WEIGHT_RECOVERY,
+    common_quantize.check_op_quantization_config,
+)
+
+register_config_check_policy_func(
+    AlgorithmName.DEQUANTIZED_WEIGHT_RECOVERY,
+    default_policy.DEFAULT_CONFIG_CHECK_POLICY,
+)
+
+DEQUANTIZED_WEIGHT_RECOVERY_OP_NAME_MATERIALIZE_FUNC_DICT = {
+    _TFLOpName.FULLY_CONNECTED: common_quantize.materialize_fc_conv,
+    _TFLOpName.CONV_2D: common_quantize.materialize_fc_conv,
+    _TFLOpName.EMBEDDING_LOOKUP: common_quantize.materialize_embedding_lookup,
+}
+
+for (
+    op_name,
+    materialize_func,
+) in DEQUANTIZED_WEIGHT_RECOVERY_OP_NAME_MATERIALIZE_FUNC_DICT.items():
+  register_quantized_op(
+      algorithm_key=AlgorithmName.DEQUANTIZED_WEIGHT_RECOVERY,
+      tfl_op_name=op_name,
+      init_qsv_func=dequantized_weight_recovery.init_qsvs,  # pyrefly: ignore[bad-argument-type]
+      calibration_func=dequantized_weight_recovery.calibrate,  # pyrefly: ignore[bad-argument-type]
+      # Most of the materialize op functions are common for all algorithms
+      # except for the function to get scale and zero point, i.e.,
+      # get_tensor_quant_params. So we use functools.partial here to pass in the
+      # common utility function and the algorithm-specific function.
+      materialize_func=functools.partial(
+          materialize_func,
+          dequantized_weight_recovery.get_tensor_quant_params,
+      ),
+  )
+
+
+# Register OCTAV algorithm.
+register_op_quant_config_validation_func(
+    AlgorithmName.OCTAV,
+    common_quantize.check_op_quantization_config,
+)
+
+# Register a config check policy for OCTAV algorithm.
+register_config_check_policy_func(
+    AlgorithmName.OCTAV,
+    default_policy.DEFAULT_CONFIG_CHECK_POLICY,
+)
+
+_OCTAV_OP_NAME_MATERIALIZE_FUNC_DICT = immutabledict({
+    _TFLOpName.INPUT: common_quantize.materialize_input,
+    _TFLOpName.OUTPUT: common_quantize.materialize_output,
+    _TFLOpName.FULLY_CONNECTED: common_quantize.materialize_fc_conv,
+    _TFLOpName.BATCH_MATMUL: common_quantize.materialize_batch_matmul,
+    _TFLOpName.CONV_2D: common_quantize.materialize_fc_conv,
+    _TFLOpName.DEPTHWISE_CONV_2D: common_quantize.materialize_fc_conv,
+    _TFLOpName.CONV_2D_TRANSPOSE: common_quantize.materialize_conv2d_transpose,
+    _TFLOpName.RESHAPE: common_quantize.materialize_reshape,
+    _TFLOpName.AVERAGE_POOL_2D: common_quantize.materialize_average_pool_2d,
+    _TFLOpName.EMBEDDING_LOOKUP: common_quantize.materialize_embedding_lookup,
+    _TFLOpName.SOFTMAX: common_quantize.materialize_softmax_and_logistic,
+    _TFLOpName.TANH: common_quantize.materialize_tanh,
+    _TFLOpName.TRANSPOSE: common_quantize.materialize_transpose,
+    _TFLOpName.GELU: common_quantize.materialize_gelu,
+    _TFLOpName.ADD: common_quantize.materialize_add,
+    _TFLOpName.SUB: common_quantize.materialize_sub,
+    _TFLOpName.MUL: common_quantize.materialize_mul,
+    _TFLOpName.MEAN: common_quantize.materialize_mean,
+    _TFLOpName.RSQRT: common_quantize.materialize_rsqrt,
+    _TFLOpName.CONCATENATION: common_quantize.materialize_concatenation,
+    _TFLOpName.STRIDED_SLICE: common_quantize.materialize_strided_slice,
+    _TFLOpName.SPLIT: common_quantize.materialize_split,
+    _TFLOpName.LOGISTIC: common_quantize.materialize_softmax_and_logistic,
+    _TFLOpName.SLICE: common_quantize.materialize_slice,
+    _TFLOpName.SUM: common_quantize.materialize_sum,
+    _TFLOpName.SELECT: common_quantize.materialize_select,
+    _TFLOpName.SELECT_V2: common_quantize.materialize_select_v2,
+    _TFLOpName.DYNAMIC_UPDATE_SLICE: (
+        common_quantize.materialize_dynamic_update_slice
+    ),
+    _TFLOpName.STABLEHLO_COMPOSITE: common_quantize.materialize_composite,
+    _TFLOpName.PAD: common_quantize.materialize_pad,
+    _TFLOpName.SQUARED_DIFFERENCE: (
+        common_quantize.materialize_squared_difference
+    ),
+    _TFLOpName.MAX_POOL_2D: common_quantize.materialize_max_pool_2d,
+    _TFLOpName.RESIZE_BILINEAR: common_quantize.materialize_resize_bilinear,
+    _TFLOpName.RESIZE_NEAREST_NEIGHBOR: (
+        common_quantize.materialize_resize_nearest_neighbor
+    ),
+    _TFLOpName.GATHER_ND: common_quantize.materialize_gather_nd,
+    _TFLOpName.PACK: common_quantize.materialize_pack,
+    _TFLOpName.UNPACK: common_quantize.materialize_unpack,
+    _TFLOpName.DIV: common_quantize.materialize_div,
+    _TFLOpName.BROADCAST_TO: common_quantize.materialize_broadcast_to,
+    _TFLOpName.SQRT: common_quantize.materialize_sqrt,
+    _TFLOpName.GATHER: common_quantize.materialize_gather,
+    _TFLOpName.HARD_SWISH: common_quantize.materialize_hard_swish,
+    _TFLOpName.MAXIMUM: common_quantize.materialize_maximum,
+    _TFLOpName.PADV2: common_quantize.materialize_padv2,
+    _TFLOpName.REDUCE_MIN: common_quantize.materialize_reduce_min,
+    _TFLOpName.EQUAL: common_quantize.materialize_equal,
+    _TFLOpName.NOT_EQUAL: common_quantize.materialize_not_equal,
+    _TFLOpName.MIRROR_PAD: common_quantize.materialize_mirror_pad,
+    _TFLOpName.SPACE_TO_DEPTH: common_quantize.materialize_space_to_depth,
+    _TFLOpName.RELU: common_quantize.materialize_relu,
+})
+
+for op_name, materialize_func in _OCTAV_OP_NAME_MATERIALIZE_FUNC_DICT.items():
+  register_quantized_op(
+      AlgorithmName.OCTAV,
+      op_name,
+      naive_min_max_quantize.init_qsvs,
+      calibration_func=naive_min_max_quantize.min_max_calibrate,
+      materialize_func=functools.partial(
+          materialize_func,
+          octav.get_tensor_quant_params,
+      ),
+  )
+
+# Register the Hadamard Rotation algorithm.
+register_op_quant_config_validation_func(
+    AlgorithmName.HADAMARD_ROTATION,
+    common_quantize.check_op_quantization_config,
+)
+
+# Register a config check policy for the Hadamard Rotation algorithm.
+register_config_check_policy_func(
+    AlgorithmName.HADAMARD_ROTATION,
+    default_policy.DEFAULT_CONFIG_CHECK_POLICY,
+)
+
+# Register specialized hadamard rotation materialize functions.
+_HADAMARD_ROTATION_OP_NAME_MATERIALIZE_FUNC_DICT = immutabledict({
+    _TFLOpName.FULLY_CONNECTED: (
+        hadamard_rotation.materialize_fully_connected_custom_op
+    ),
+    _TFLOpName.EMBEDDING_LOOKUP: (
+        hadamard_rotation.materialize_embedding_lookup_custom_op
+    ),
+})
+for (
+    op_name,
+    materialize_func,
+) in _HADAMARD_ROTATION_OP_NAME_MATERIALIZE_FUNC_DICT.items():
+  register_quantized_op(
+      AlgorithmName.HADAMARD_ROTATION,
+      op_name,
+      naive_min_max_quantize.init_qsvs,
+      calibration_func=naive_min_max_quantize.min_max_calibrate,
+      materialize_func=materialize_func,  # pyrefly: ignore[bad-argument-type]
+  )
+
+register_op_quant_config_validation_func(
+    AlgorithmName.DECOMPOSED_HADAMARD_ROTATION,
+    common_quantize.check_op_quantization_config,
+)
+
+register_config_check_policy_func(
+    AlgorithmName.DECOMPOSED_HADAMARD_ROTATION,
+    default_policy.DEFAULT_CONFIG_CHECK_POLICY,
+)
+
+_DECOMPOSED_HADAMARD_ROTATION_OP_NAME_MATERIALIZE_FUNC_DICT = immutabledict({
+    _TFLOpName.FULLY_CONNECTED: (
+        hadamard_rotation.materialize_fully_connected_decomposed
+    ),
+    _TFLOpName.EMBEDDING_LOOKUP: (
+        hadamard_rotation.materialize_embedding_lookup_decomposed
+    ),
+})
+for (
+    op_name,
+    materialize_func,
+) in _DECOMPOSED_HADAMARD_ROTATION_OP_NAME_MATERIALIZE_FUNC_DICT.items():
+  register_quantized_op(
+      AlgorithmName.DECOMPOSED_HADAMARD_ROTATION,
+      op_name,
+      naive_min_max_quantize.init_qsvs,
+      calibration_func=naive_min_max_quantize.min_max_calibrate,
+      materialize_func=materialize_func,  # pyrefly: ignore[bad-argument-type]
+  )
+
+
+# Register the MSE algorithm.
+register_op_quant_config_validation_func(
+    AlgorithmName.MSE,
+    common_quantize.check_op_quantization_config,
+)
+
+# Register a config check policy for the MSE algorithm.
+register_config_check_policy_func(
+    AlgorithmName.MSE,
+    default_policy.DEFAULT_CONFIG_CHECK_POLICY,
+)
+
+# Register specialized MSE materialize functions.
+_MSE_OP_NAME_MATERIALIZE_FUNC_DICT = immutabledict({
+    _TFLOpName.FULLY_CONNECTED: common_quantize.materialize_fc_conv,
+    _TFLOpName.EMBEDDING_LOOKUP: common_quantize.materialize_embedding_lookup,
+    _TFLOpName.CONV_2D: common_quantize.materialize_fc_conv,
+    _TFLOpName.DEPTHWISE_CONV_2D: common_quantize.materialize_fc_conv,
+    _TFLOpName.CONV_2D_TRANSPOSE: common_quantize.materialize_conv2d_transpose,
+})
+for (
+    op_name,
+    materialize_func,
+) in _MSE_OP_NAME_MATERIALIZE_FUNC_DICT.items():
+  register_quantized_op(
+      AlgorithmName.MSE,
+      op_name,
+      naive_min_max_quantize.init_qsvs,
+      calibration_func=naive_min_max_quantize.min_max_calibrate,
+      materialize_func=functools.partial(
+          materialize_func,
+          mse.get_tensor_quant_params,
+      ),
+  )
+
+# Register the GPTQ algorithm.
+register_op_quant_config_validation_func(
+    AlgorithmName.GPTQ,
+    common_quantize.check_op_quantization_config,
+)
+
+# Register a config check policy for the GPTQ algorithm.
+register_config_check_policy_func(
+    AlgorithmName.GPTQ,
+    default_policy.DEFAULT_CONFIG_CHECK_POLICY,
+)
+
+# Register specialized GPTQ materialize functions.
+_GPTQ_OP_NAME_MATERIALIZE_FUNC_DICT = immutabledict({
+    _TFLOpName.FULLY_CONNECTED: common_quantize.materialize_fc_conv,
+})
+for (
+    op_name,
+    materialize_func,
+) in _GPTQ_OP_NAME_MATERIALIZE_FUNC_DICT.items():
+  register_quantized_op(
+      AlgorithmName.GPTQ,
+      op_name,
+      naive_min_max_quantize.init_qsvs,
+      calibration_func=gptq.calibrate,  # pyrefly: ignore[bad-argument-type]
+      materialize_func=functools.partial(
+          materialize_func,
+          gptq.get_tensor_quant_params,
+      ),
+      update_qsv_func=qsv_utils.gptq_and_moving_average_update,  # pyrefly: ignore[bad-argument-type]
+  )
+
+# Register the OSCAR algorithm.
+register_op_quant_config_validation_func(
+    AlgorithmName.OSCAR,
+    common_quantize.check_op_quantization_config,
+)
+
+# Register a config check policy for the OSCAR algorithm.
+register_config_check_policy_func(
+    AlgorithmName.OSCAR,
+    default_policy.DEFAULT_CONFIG_CHECK_POLICY,
+)
+
+# Register specialized OSCAR materialize functions.
+_OSCAR_OP_NAME_MATERIALIZE_FUNC_DICT = immutabledict({
+    _TFLOpName.FULLY_CONNECTED: oscar.materialize_fully_connected,
+})
+for (
+    op_name,
+    materialize_func,
+) in _OSCAR_OP_NAME_MATERIALIZE_FUNC_DICT.items():
+  register_quantized_op(
+      AlgorithmName.OSCAR,
+      op_name,
+      naive_min_max_quantize.init_qsvs,
+      calibration_func=oscar.calibrate,  # pyrefly: ignore[bad-argument-type]
+      materialize_func=materialize_func,  # pyrefly: ignore[bad-argument-type]
+      update_qsv_func=qsv_utils.oscar_and_moving_average_update,  # pyrefly: ignore[bad-argument-type]
+  )