neural-compressor 3.3__tar.gz → 3.4__tar.gz

{neural_compressor-3.3 → neural_compressor-3.4}/PKG-INFO

@@ -1,6 +1,6 @@
-Metadata-Version: 2.2
+Metadata-Version: 2.1
 Name: neural_compressor
-Version: 3.3
+Version: 3.4
 Summary: Repository of Intel® Neural Compressor
 Home-page: https://github.com/intel/neural-compressor
 Author: Intel AIPT Team
@@ -30,8 +30,7 @@ Requires-Dist: requests
 Requires-Dist: schema
 Requires-Dist: scikit-learn
 Provides-Extra: pt
-Requires-Dist: numpy==1.23.5; python_version < "3.12" and extra == "pt"
-Requires-Dist: numpy<2.0; python_version >= "3.12" and extra == "pt"
+Requires-Dist: numpy; extra == "pt"
 Requires-Dist: prettytable; extra == "pt"
 Requires-Dist: psutil; extra == "pt"
 Requires-Dist: py-cpuinfo; extra == "pt"
@@ -43,18 +42,6 @@ Requires-Dist: py-cpuinfo; extra == "tf"
 Requires-Dist: pydantic; extra == "tf"
 Requires-Dist: pyyaml; extra == "tf"
 Requires-Dist: tensorflow; extra == "tf"
-Dynamic: author
-Dynamic: author-email
-Dynamic: classifier
-Dynamic: description
-Dynamic: description-content-type
-Dynamic: home-page
-Dynamic: keywords
-Dynamic: license
-Dynamic: provides-extra
-Dynamic: requires-dist
-Dynamic: requires-python
-Dynamic: summary
 
 <div align="center">
 
@@ -63,7 +50,7 @@ Intel® Neural Compressor
 <h3> An open-source Python library supporting popular model compression techniques on all mainstream deep learning frameworks (TensorFlow, PyTorch, and ONNX Runtime)</h3>
 
 [![python](https://img.shields.io/badge/python-3.8%2B-blue)](https://github.com/intel/neural-compressor)
-[![version](https://img.shields.io/badge/release-3.3-green)](https://github.com/intel/neural-compressor/releases)
+[![version](https://img.shields.io/badge/release-3.4-green)](https://github.com/intel/neural-compressor/releases)
 [![license](https://img.shields.io/badge/license-Apache%202-blue)](https://github.com/intel/neural-compressor/blob/master/LICENSE)
 [![coverage](https://img.shields.io/badge/coverage-85%25-green)](https://github.com/intel/neural-compressor)
 [![Downloads](https://static.pepy.tech/personalized-badge/neural-compressor?period=total&units=international_system&left_color=grey&right_color=green&left_text=downloads)](https://pepy.tech/project/neural-compressor)
@@ -93,7 +80,7 @@ support AMD CPU, ARM CPU, and NVidia GPU through ONNX Runtime with limited testi
 Choose the necessary framework dependencies to install based on your deploy environment.
 ### Install Framework
 * [Install intel_extension_for_pytorch for CPU](https://intel.github.io/intel-extension-for-pytorch/cpu/latest/)    
-* [Install intel_extension_for_pytorch for XPU](https://intel.github.io/intel-extension-for-pytorch/xpu/latest/)    
+* [Install intel_extension_for_pytorch for Intel GPU](https://intel.github.io/intel-extension-for-pytorch/xpu/latest/)    
 * [Use Docker Image with torch installed for HPU](https://docs.habana.ai/en/latest/Installation_Guide/Bare_Metal_Fresh_OS.html#bare-metal-fresh-os-single-click)    
   **Note**: There is a version mapping between Intel Neural Compressor and Gaudi Software Stack, please refer to this [table](./docs/source/3x/gaudi_version_map.md) and make sure to use a matched combination.    
 * [Install torch for other platform](https://pytorch.org/get-started/locally)    
@@ -114,8 +101,11 @@ To try on Intel Gaudi2, docker image with Gaudi Software Stack is recommended, p
 
 Run a container with an interactive shell, [more info](https://docs.habana.ai/en/latest/Installation_Guide/Additional_Installation/Docker_Installation.html#docker-installation)
 ```
-docker run -it --runtime=habana -e HABANA_VISIBLE_DEVICES=all -e OMPI_MCA_btl_vader_single_copy_mechanism=none --cap-add=sys_nice --net=host --ipc=host vault.habana.ai/gaudi-docker/1.20.0/ubuntu24.04/habanalabs/pytorch-installer-2.6.0:latest
+docker run -it --runtime=habana -e HABANA_VISIBLE_DEVICES=all -e OMPI_MCA_btl_vader_single_copy_mechanism=none --cap-add=sys_nice --net=host --ipc=host vault.habana.ai/gaudi-docker/1.21.0/ubuntu24.04/habanalabs/pytorch-installer-2.6.0:latest
 ```
+
+> Note: Since Habana software >= 1.21.0, `PT_HPU_LAZY_MODE=0` is the default setting. However, most low-precision functions (such as `convert_from_uint4`) do not support this setting. Therefore, we recommend setting `PT_HPU_LAZY_MODE=1` to maintain compatibility.
+
 Run the example,
 ```python
 from neural_compressor.torch.quantization import (
@@ -231,12 +221,10 @@ model = load(
 
 ## Selected Publications/Events
 
+* arXiv: [Faster Inference of LLMs using FP8 on the Intel Gaudi](https://arxiv.org/abs/2503.09975) (Mar 2025)
+* PyTorch landscape: [PyTorch general optimizations](https://landscape.pytorch.org/) (Mar 2025)
+* Blog on SqueezeBits: [[Intel Gaudi] #4. FP8 Quantization](https://blog.squeezebits.com/intel-gaudi-4-fp8-quantization--40269) (Jan 2025)
 * EMNLP'2024: [Optimize Weight Rounding via Signed Gradient Descent for the Quantization of LLMs](https://arxiv.org/abs/2309.05516) (Sep 2024)
-* Blog on Medium: [Quantization on Intel Gaudi Series AI Accelerators](https://medium.com/intel-analytics-software/intel-neural-compressor-v3-0-a-quantization-tool-across-intel-hardware-9856adee6f11) (Aug 2024)
-* Blog by Intel: [Neural Compressor: Boosting AI Model Efficiency](https://community.intel.com/t5/Blogs/Tech-Innovation/Artificial-Intelligence-AI/Neural-Compressor-Boosting-AI-Model-Efficiency/post/1604740) (June 2024)
-* Blog by Intel: [Optimization of Intel AI Solutions for Alibaba Cloud’s Qwen2 Large Language Models](https://www.intel.com/content/www/us/en/developer/articles/technical/intel-ai-solutions-accelerate-alibaba-qwen2-llms.html) (June 2024)
-* Blog by Intel: [Accelerate Meta* Llama 3 with Intel AI Solutions](https://www.intel.com/content/www/us/en/developer/articles/technical/accelerate-meta-llama3-with-intel-ai-solutions.html) (Apr 2024)
-* EMNLP'2023 (Under Review): [TEQ: Trainable Equivalent Transformation for Quantization of LLMs](https://openreview.net/forum?id=iaI8xEINAf&referrer=%5BAuthor%20Console%5D) (Sep 2023)
 * arXiv: [Efficient Post-training Quantization with FP8 Formats](https://arxiv.org/abs/2309.14592) (Sep 2023)
 * arXiv: [Optimize Weight Rounding via Signed Gradient Descent for the Quantization of LLMs](https://arxiv.org/abs/2309.05516) (Sep 2023)
 

{neural_compressor-3.3 → neural_compressor-3.4}/README.md

@@ -5,7 +5,7 @@ Intel® Neural Compressor
 <h3> An open-source Python library supporting popular model compression techniques on all mainstream deep learning frameworks (TensorFlow, PyTorch, and ONNX Runtime)</h3>
 
 [![python](https://img.shields.io/badge/python-3.8%2B-blue)](https://github.com/intel/neural-compressor)
-[![version](https://img.shields.io/badge/release-3.3-green)](https://github.com/intel/neural-compressor/releases)
+[![version](https://img.shields.io/badge/release-3.4-green)](https://github.com/intel/neural-compressor/releases)
 [![license](https://img.shields.io/badge/license-Apache%202-blue)](https://github.com/intel/neural-compressor/blob/master/LICENSE)
 [![coverage](https://img.shields.io/badge/coverage-85%25-green)](https://github.com/intel/neural-compressor)
 [![Downloads](https://static.pepy.tech/personalized-badge/neural-compressor?period=total&units=international_system&left_color=grey&right_color=green&left_text=downloads)](https://pepy.tech/project/neural-compressor)
@@ -35,7 +35,7 @@ support AMD CPU, ARM CPU, and NVidia GPU through ONNX Runtime with limited testi
 Choose the necessary framework dependencies to install based on your deploy environment.
 ### Install Framework
 * [Install intel_extension_for_pytorch for CPU](https://intel.github.io/intel-extension-for-pytorch/cpu/latest/)    
-* [Install intel_extension_for_pytorch for XPU](https://intel.github.io/intel-extension-for-pytorch/xpu/latest/)    
+* [Install intel_extension_for_pytorch for Intel GPU](https://intel.github.io/intel-extension-for-pytorch/xpu/latest/)    
 * [Use Docker Image with torch installed for HPU](https://docs.habana.ai/en/latest/Installation_Guide/Bare_Metal_Fresh_OS.html#bare-metal-fresh-os-single-click)    
   **Note**: There is a version mapping between Intel Neural Compressor and Gaudi Software Stack, please refer to this [table](./docs/source/3x/gaudi_version_map.md) and make sure to use a matched combination.    
 * [Install torch for other platform](https://pytorch.org/get-started/locally)    
@@ -56,8 +56,11 @@ To try on Intel Gaudi2, docker image with Gaudi Software Stack is recommended, p
 
 Run a container with an interactive shell, [more info](https://docs.habana.ai/en/latest/Installation_Guide/Additional_Installation/Docker_Installation.html#docker-installation)
 ```
-docker run -it --runtime=habana -e HABANA_VISIBLE_DEVICES=all -e OMPI_MCA_btl_vader_single_copy_mechanism=none --cap-add=sys_nice --net=host --ipc=host vault.habana.ai/gaudi-docker/1.20.0/ubuntu24.04/habanalabs/pytorch-installer-2.6.0:latest
+docker run -it --runtime=habana -e HABANA_VISIBLE_DEVICES=all -e OMPI_MCA_btl_vader_single_copy_mechanism=none --cap-add=sys_nice --net=host --ipc=host vault.habana.ai/gaudi-docker/1.21.0/ubuntu24.04/habanalabs/pytorch-installer-2.6.0:latest
 ```
+
+> Note: Since Habana software >= 1.21.0, `PT_HPU_LAZY_MODE=0` is the default setting. However, most low-precision functions (such as `convert_from_uint4`) do not support this setting. Therefore, we recommend setting `PT_HPU_LAZY_MODE=1` to maintain compatibility.
+
 Run the example,
 ```python
 from neural_compressor.torch.quantization import (
@@ -173,12 +176,10 @@ model = load(
 
 ## Selected Publications/Events
 
+* arXiv: [Faster Inference of LLMs using FP8 on the Intel Gaudi](https://arxiv.org/abs/2503.09975) (Mar 2025)
+* PyTorch landscape: [PyTorch general optimizations](https://landscape.pytorch.org/) (Mar 2025)
+* Blog on SqueezeBits: [[Intel Gaudi] #4. FP8 Quantization](https://blog.squeezebits.com/intel-gaudi-4-fp8-quantization--40269) (Jan 2025)
 * EMNLP'2024: [Optimize Weight Rounding via Signed Gradient Descent for the Quantization of LLMs](https://arxiv.org/abs/2309.05516) (Sep 2024)
-* Blog on Medium: [Quantization on Intel Gaudi Series AI Accelerators](https://medium.com/intel-analytics-software/intel-neural-compressor-v3-0-a-quantization-tool-across-intel-hardware-9856adee6f11) (Aug 2024)
-* Blog by Intel: [Neural Compressor: Boosting AI Model Efficiency](https://community.intel.com/t5/Blogs/Tech-Innovation/Artificial-Intelligence-AI/Neural-Compressor-Boosting-AI-Model-Efficiency/post/1604740) (June 2024)
-* Blog by Intel: [Optimization of Intel AI Solutions for Alibaba Cloud’s Qwen2 Large Language Models](https://www.intel.com/content/www/us/en/developer/articles/technical/intel-ai-solutions-accelerate-alibaba-qwen2-llms.html) (June 2024)
-* Blog by Intel: [Accelerate Meta* Llama 3 with Intel AI Solutions](https://www.intel.com/content/www/us/en/developer/articles/technical/accelerate-meta-llama3-with-intel-ai-solutions.html) (Apr 2024)
-* EMNLP'2023 (Under Review): [TEQ: Trainable Equivalent Transformation for Quantization of LLMs](https://openreview.net/forum?id=iaI8xEINAf&referrer=%5BAuthor%20Console%5D) (Sep 2023)
 * arXiv: [Efficient Post-training Quantization with FP8 Formats](https://arxiv.org/abs/2309.14592) (Sep 2023)
 * arXiv: [Optimize Weight Rounding via Signed Gradient Descent for the Quantization of LLMs](https://arxiv.org/abs/2309.05516) (Sep 2023)
 

{neural_compressor-3.3 → neural_compressor-3.4}/neural_compressor/__init__.py

@@ -17,16 +17,18 @@
 """Intel® Neural Compressor: An open-source Python library supporting popular model compression techniques."""
 from .version import __version__
 
-# we need to set a global 'NA' backend, or Model can't be used
-from .config import (
-    DistillationConfig,
-    PostTrainingQuantConfig,
-    WeightPruningConfig,
-    QuantizationAwareTrainingConfig,
-    MixedPrecisionConfig,
-)
-from .contrib import *
-from .model import *
-from .metric import *
-from .utils import options
-from .utils.utility import set_random_seed, set_tensorboard, set_workspace, set_resume_from
+import os
+
+if not (os.environ.get("INC_PT_ONLY", False) or os.environ.get("INC_TF_ONLY", False)):
+    from .config import (
+        DistillationConfig,
+        PostTrainingQuantConfig,
+        WeightPruningConfig,
+        QuantizationAwareTrainingConfig,
+        MixedPrecisionConfig,
+    )
+    from .contrib import *
+    from .model import *
+    from .metric import *
+    from .utils import options
+    from .utils.utility import set_random_seed, set_tensorboard, set_workspace, set_resume_from

{neural_compressor-3.3 → neural_compressor-3.4}/neural_compressor/adaptor/adaptor.py

@@ -17,7 +17,7 @@
 
 from abc import abstractmethod
 
-"""The framework backends supported by neural_compressor, including tensorflow, mxnet and pytorch.
+"""The framework backends supported by neural_compressor, including tensorflow and pytorch.
 
    User could add new backend support by implementing new Adaptor subclass under this directory.
    The naming convention of new Adaptor subclass should be something like ABCAdaptor, user

{neural_compressor-3.3 → neural_compressor-3.4}/neural_compressor/adaptor/keras.py

@@ -23,6 +23,7 @@ from collections import OrderedDict, UserDict
 
 import numpy as np
 import yaml
+from deprecated import deprecated
 
 from ..data.dataloaders.base_dataloader import BaseDataLoader
 from ..utils import logger
@@ -68,6 +69,7 @@ def _add_supported_quantized_objects(custom_objects):
     return custom_objects
 
 
+@deprecated(reason="KerasAdaptor is deprecated and may be removed in future versions.", version="3.4")
 @adaptor_registry
 class KerasAdaptor(Adaptor):
     """The keras class of framework adaptor layer."""

{neural_compressor-3.3 → neural_compressor-3.4}/neural_compressor/adaptor/onnxrt.py

@@ -30,6 +30,7 @@ from typing import Dict
 
 import numpy as np
 import yaml
+from deprecated import deprecated
 from packaging.version import Version
 
 from neural_compressor.adaptor.adaptor import Adaptor, adaptor_registry
@@ -48,6 +49,7 @@ ONNXRT112_VERSION = Version("1.12.0")
 logger = logging.getLogger("neural_compressor")
 
 
+@deprecated(reason="ONNXRUNTIMEAdaptor is deprecated and may be removed in future versions.", version="3.4")
 @adaptor_registry
 class ONNXRUNTIMEAdaptor(Adaptor):
     """The ONNXRT adaptor layer, do onnx-rt quantization, calibration, inspect layer tensors.

{neural_compressor-3.3 → neural_compressor-3.4}/neural_compressor/adaptor/pytorch.py

@@ -4170,8 +4170,12 @@ class PyTorch_FXAdaptor(TemplateAdaptor):
                     sub_name = node.target
                 if not hasattr(model, node.target):
                     continue
-                if "scale" in node.target:
-                    tune_cfg["get_attr"][sub_name] = float(getattr(model, node.target))
+                # Improved scale detection logic
+                if "scale" in node.target and not any(exclude in node.target for exclude in ["layer_scale", "gamma"]):
+                    try:
+                        tune_cfg["get_attr"][sub_name] = getattr(model, node.target).tolist()
+                    except Exception as e:
+                        logger.warning(f"Could not convert {node.target} to list, skipping... Error: {str(e)}")
                 elif "zero_point" in node.target:
                     tune_cfg["get_attr"][sub_name] = int(getattr(model, node.target))
                 else:

{neural_compressor-3.3 → neural_compressor-3.4}/neural_compressor/adaptor/tensorflow.py

@@ -23,6 +23,7 @@ from collections import OrderedDict, UserDict
 
 import numpy as np
 import yaml
+from deprecated import deprecated
 
 from ..data.dataloaders.base_dataloader import BaseDataLoader
 from ..utils import logger
@@ -55,6 +56,7 @@ spr_base_verions = (
 )
 
 
+@deprecated(reason="TensorFlowAdaptor is deprecated and may be removed in future versions.", version="3.4")
 @adaptor_registry
 class TensorFlowAdaptor(Adaptor):
     """Adaptor Layer for stock tensorflow and spr-base."""

{neural_compressor-3.3 → neural_compressor-3.4}/neural_compressor/adaptor/torch_utils/layer_wise_quant/modified_pickle.py

@@ -483,7 +483,7 @@ class _Pickler:  # pragma: no cover
         The memo is the data structure that remembers which objects the
         pickler has already seen, so that shared or recursive objects
         are pickled by reference and not by value.  This method is
-        useful when re-using picklers.
+        useful when reusing picklers.
         """
         self.memo.clear()
 

{neural_compressor-3.3 → neural_compressor-3.4}/neural_compressor/benchmark.py

@@ -166,8 +166,6 @@ def run_instance(model, conf, b_dataloader=None, b_func=None):
             )
         if framework == "keras":
             framework_specific_info.update({"workspace_path": options.workspace})
-        if framework == "mxnet":
-            framework_specific_info.update({"b_dataloader": b_dataloader})
         if "onnx" in framework:
             framework_specific_info.update(
                 {"workspace_path": options.workspace, "graph_optimization": OPTIONS[framework].graph_optimization}

{neural_compressor-3.3 → neural_compressor-3.4}/neural_compressor/common/base_config.py

@@ -189,6 +189,7 @@ class BaseConfig(ABC):
 
     name = BASE_CONFIG
     params_list = []
+    _is_initialized = False
 
     def __init__(self, white_list: Optional[List[OP_NAME_OR_MODULE_TYPE]] = DEFAULT_WHITE_LIST) -> None:
         """Initialize the BaseConfig.
@@ -220,6 +221,14 @@ class BaseConfig(ABC):
                 f"The white list should be one of {DEFAULT_WHITE_LIST}, {EMPTY_WHITE_LIST},"
                 " a not empty list, but got {self.white_list}"
             )
+        self._is_initialized = True
+
+    def __setattr__(self, name, value):
+        """Override the setattr function to propagate updates."""
+        super().__setattr__(name, value)
+        if self._is_initialized and name in self.params_list:
+            self._is_initialized = False
+            self._post_init()
 
     @property
     def white_list(self):
@@ -683,6 +692,13 @@ class ComposableConfig(BaseConfig):
             self.config_list.append(other)
         return self
 
+    def __setattr__(self, name, value):
+        """Override the setattr function to propagate updates."""
+        ABC.__setattr__(self, name, value)
+        for config in self.config_list:
+            if hasattr(config, name):
+                setattr(config, name, value)
+
     def to_dict(self, params_list=[], operator2str=None):
         """Converts the configuration object to a dictionary.
 
@@ -884,7 +900,6 @@ class Options:
 
     def __init__(self, random_seed=1978, workspace=DEFAULT_WORKSPACE, resume_from=None, tensorboard=False):
         """Init an Option object."""
-        os.makedirs(workspace, exist_ok=True)
         self.random_seed = random_seed
         self.workspace = workspace
         self.resume_from = resume_from

{neural_compressor-3.3 → neural_compressor-3.4}/neural_compressor/common/utils/constants.py

@@ -34,6 +34,7 @@ HQQ = "hqq"  # pragma: no cover
 TEQ = "teq"  # pragma: no cover
 AUTOROUND = "autoround"
 FP8_QUANT = "fp8_quant"
+HYBRID_GPTQ = "hybrid_gptq"
 MX_QUANT = "mx_quant"
 MIXED_PRECISION = "mixed_precision"
 
@@ -51,12 +52,13 @@ from enum import Enum
 
 
 class Mode(Enum):
-    """Enumeration class representing different modes of the quantizer execution."""
+    """Enumeration class representing different modes of the quantization."""
 
     PREPARE = "prepare"
     CONVERT = "convert"
     QUANTIZE = "quantize"
     LOAD = "load"
+    SAVE = "save"
 
 
 SERVER_PROCESSOR_BRAND_KEY_WORLD_LST = ["Xeon"]

{neural_compressor-3.3 → neural_compressor-3.4}/neural_compressor/common/utils/logger.py

@@ -161,6 +161,8 @@ def _get_log_msg(mode):
         log_msg = "Conversion"
     elif mode == Mode.LOAD:  # pragma: no cover
         log_msg = "Loading"
+    elif mode == Mode.SAVE:  # pragma: no cover
+        log_msg = "Saving"
     return log_msg
 
 

{neural_compressor-3.3 → neural_compressor-3.4}/neural_compressor/common/utils/save_load.py

@@ -47,11 +47,23 @@ def load_config_mapping(qconfig_file_path, config_name_mapping):  # pragma: no c
     Returns:
         config_mapping (dict): config mapping.
     """
+
+    def _fetch_from_string(key):
+        """Return op_name and op_type from key, such as "('transformer.h.0.attn.k_proj', 'Linear')"."""
+        import re
+
+        match = re.match(r"\('(.+)', '(.+)'\)", key)
+        if match:
+            op_name, op_type = match.groups()
+            return op_name, op_type
+        else:
+            raise ValueError(f"Invalid key format: {key}. Expected format: \"('op_name', 'op_type')\".")
+
     config_mapping = {}
     with open(qconfig_file_path, "r") as f:
         per_op_qconfig = json.load(f)
     for key, value in per_op_qconfig.items():
-        op_name, op_type = eval(key)
+        op_name, op_type = _fetch_from_string(key)
         # value here is a dict, so we convert it to an object with config_name_mapping,
         # which is defined in a specific framework.
         config_name = next(iter(value))

{neural_compressor-3.3 → neural_compressor-3.4}/neural_compressor/common/utils/utility.py

@@ -108,13 +108,13 @@ class CpuInfo(object):
             max_extension_support = cpuid.get_max_extension_support()
             if max_extension_support >= 7:
                 ecx = cpuid._run_asm(
-                    b"\x31\xC9",  # xor ecx, ecx
-                    b"\xB8\x07\x00\x00\x00" b"\x0f\xa2" b"\x89\xC8" b"\xC3",  # mov eax, 7  # cpuid  # mov ax, cx  # ret
+                    b"\x31\xc9",  # xor ecx, ecx
+                    b"\xb8\x07\x00\x00\x00" b"\x0f\xa2" b"\x89\xc8" b"\xc3",  # mov eax, 7  # cpuid  # mov ax, cx  # ret
                 )
                 self._vnni = bool(ecx & (1 << 11))
                 eax = cpuid._run_asm(
-                    b"\xB9\x01\x00\x00\x00",  # mov ecx, 1
-                    b"\xB8\x07\x00\x00\x00" b"\x0f\xa2" b"\xC3",  # mov eax, 7  # cpuid  # ret
+                    b"\xb9\x01\x00\x00\x00",  # mov ecx, 1
+                    b"\xb8\x07\x00\x00\x00" b"\x0f\xa2" b"\xc3",  # mov eax, 7  # cpuid  # ret
                 )
                 self._bf16 = bool(eax & (1 << 5))
         self._info = info

{neural_compressor-3.3 → neural_compressor-3.4}/neural_compressor/common/version.py

@@ -15,4 +15,4 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """Intel® Neural Compressor: An open-source Python library supporting popular model compression techniques."""
-__version__ = "3.3"
+__version__ = "3.4"

{neural_compressor-3.3 → neural_compressor-3.4}/neural_compressor/config.py

@@ -2361,11 +2361,11 @@ class NASConfig:
         self._search = search
 
 
-class MXNet:
-    """Base config class for MXNet."""
+class PyTorch:
+    """Base config class for PyTorch."""
 
     def __init__(self, precisions=None):
-        """Init an MXNet object."""
+        """Init an PyTorch object."""
         self._precisions = precisions
 
     @property
@@ -2383,7 +2383,7 @@ class MXNet:
         self._precisions = precisions
 
 
-class ONNX(MXNet):
+class ONNX(PyTorch):
     """Config class for ONNX."""
 
     def __init__(self, graph_optimization_level=None, precisions=None):
@@ -2408,7 +2408,7 @@ class ONNX(MXNet):
             self._graph_optimization_level = graph_optimization_level
 
 
-class TensorFlow(MXNet):
+class TensorFlow(PyTorch):
     """Config class for TensorFlow."""
 
     def __init__(self, precisions=None):
@@ -2416,7 +2416,7 @@ class TensorFlow(MXNet):
         super().__init__(precisions)
 
 
-class Keras(MXNet):
+class Keras(PyTorch):
     """Config class for Keras."""
 
     def __init__(self, precisions=None):
@@ -2424,14 +2424,6 @@ class Keras(MXNet):
         super().__init__(precisions)
 
 
-class PyTorch(MXNet):
-    """Config class for PyTorch."""
-
-    def __init__(self, precisions=None):
-        """Init a PyTorch object."""
-        super().__init__(precisions)
-
-
 quantization = PostTrainingQuantConfig()
 benchmark = BenchmarkConfig()
 options = Options()
@@ -2443,7 +2435,6 @@ onnxruntime_config = ONNX()
 tensorflow_config = TensorFlow()
 keras_config = Keras()
 pytorch_config = PyTorch()
-mxnet_config = MXNet()
 
 
 class _Config:
@@ -2460,7 +2451,6 @@ class _Config:
         onnxruntime=onnxruntime_config,
         tensorflow=tensorflow_config,
         pytorch=pytorch_config,
-        mxnet=mxnet_config,
         keras=keras_config,
     ):
         """Init a config object."""
@@ -2473,7 +2463,6 @@ class _Config:
         self._nas = nas
         self._tensorflow = tensorflow
         self._pytorch = pytorch
-        self._mxnet = mxnet
         self._keras = keras
 
     @property
@@ -2501,11 +2490,6 @@ class _Config:
         """Get the pytorch object."""
         return self._pytorch
 
-    @property
-    def mxnet(self):
-        """Get the mxnet object."""
-        return self._mxnet
-
     @property
     def pruning(self):
         """Get the pruning object."""

{neural_compressor-3.3 → neural_compressor-3.4}/neural_compressor/data/dataloaders/dataloader.py

@@ -15,7 +15,6 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """Built-in dataloaders for multiple framework backends."""
-from .mxnet_dataloader import MXNetDataLoader
 from .onnxrt_dataloader import ONNXRTDataLoader
 from .pytorch_dataloader import PyTorchDataLoader
 from .tensorflow_dataloader import TensorflowDataLoader
@@ -24,7 +23,6 @@ DATALOADERS = {
     "tensorflow": TensorflowDataLoader,
     "tensorflow_itex": TensorflowDataLoader,
     "keras": TensorflowDataLoader,
-    "mxnet": MXNetDataLoader,
     "pytorch": PyTorchDataLoader,
     "pytorch_ipex": PyTorchDataLoader,
     "pytorch_fx": PyTorchDataLoader,
@@ -89,8 +87,7 @@ class DataLoader(object):
             "onnxrt_qdqops",
             "onnxrt_qlinearops",
             "onnxrt_integerops",
-            "mxnet",
-        ), "framework support tensorflow pytorch mxnet onnxruntime"
+        ), "framework support tensorflow pytorch onnxruntime"
         return DATALOADERS[framework](
             dataset=dataset,
             batch_size=batch_size,

{neural_compressor-3.3 → neural_compressor-3.4}/neural_compressor/data/dataloaders/tensorflow_dataloader.py

@@ -41,7 +41,7 @@ class TFDataDataLoader(BaseDataLoader):  # pragma: no cover
 
     In tensorflow1.x dataloader is coupled with the graph, but it also support feed_dict
     method to do session run, this dataloader is designed to satisfy the usage of feed dict
-    in tf1.x. Although it's a general dataloader and can be used in MXNet and PyTorch.
+    in tf1.x. Although it's a general dataloader and can be used in PyTorch.
 
     Args:
         dataset: obj. wrapper of needed data.

{neural_compressor-3.3 → neural_compressor-3.4}/neural_compressor/data/datasets/coco_dataset.py

@@ -38,7 +38,6 @@ from neural_compressor.utils.utility import LazyImport
 from .dataset import Dataset, IterableDataset, dataset_registry
 
 tf = LazyImport("tensorflow")
-mx = LazyImport("mxnet")
 torch = LazyImport("torch")
 
 
@@ -160,7 +159,7 @@ class COCORecordDataset(IterableDataset):  # pragma: no cover
 @dataset_registry(
     dataset_type="COCORaw",
     framework="onnxrt_qlinearops, \
-                    onnxrt_integerops, pytorch, mxnet, tensorflow, \
+                    onnxrt_integerops, pytorch, tensorflow, \
                     tensorflow_itex",
     dataset_format="",
 )
@@ -263,7 +262,7 @@ class COCORaw(Dataset):  # pragma: no cover
 @dataset_registry(
     dataset_type="COCONpy",
     framework="onnxrt_qlinearops, \
-                    onnxrt_integerops, pytorch, mxnet, tensorflow, \
+                    onnxrt_integerops, pytorch, tensorflow, \
                     tensorflow_itex",
     dataset_format="",
 )