[PTQ][OV] FP8 implementation (openvinotoolkit#2283)

### Changes

- Added FP8 implementation
- Added `Mode` parameter

### Reason for changes

- New FP8 implementation

### Related tickets

- 119805

### Tests

- `tests/openvino/native/quantization/test_graphs.py`
- `tests/openvino/native/test_model_transformer.py`

On top of openvinotoolkit/openvino#21034 -
**Merged**

nncf/__init__.py

@@ -21,6 +21,7 @@
 from nncf.parameters import CompressWeightsMode as CompressWeightsMode
 from nncf.parameters import DropType as DropType
 from nncf.parameters import ModelType as ModelType
+from nncf.parameters import QuantizationMode as QuantizationMode
 from nncf.parameters import TargetDevice as TargetDevice
 from nncf.quantization import QuantizationPreset as QuantizationPreset
 from nncf.quantization import compress_weights as compress_weights

nncf/common/hardware/config.py

@@ -20,7 +20,7 @@
 from nncf.common.graph.operator_metatypes import OperatorMetatype
 from nncf.common.logging import nncf_logger
 from nncf.common.quantization import quantizers as quant
-from nncf.common.quantization.structs import QuantizationMode
+from nncf.common.quantization.structs import QuantizationScheme as QuantizationMode
 from nncf.common.quantization.structs import QuantizerConfig
 from nncf.common.utils.helpers import product_dict
 from nncf.common.utils.os import safe_open

nncf/common/quantization/initialization/range.py

@@ -12,7 +12,7 @@
 from typing import Dict, List, Optional
 
 from nncf.common.initialization.dataloader import NNCFDataLoader
-from nncf.common.quantization.structs import QuantizationMode
+from nncf.common.quantization.structs import QuantizationScheme as QuantizationMode
 from nncf.common.quantization.structs import QuantizerGroup
 from nncf.config.schemata.defaults import NUM_INIT_SAMPLES
 

nncf/common/quantization/quantizer_propagation/graph.py

@@ -42,7 +42,7 @@
 from nncf.common.quantization.quantizer_setup import QuantizationInsertionPointBase
 from nncf.common.quantization.quantizer_setup import QuantizationPointId
 from nncf.common.quantization.quantizer_setup import WeightQuantizationInsertionPoint
-from nncf.common.quantization.structs import QuantizationMode
+from nncf.common.quantization.structs import QuantizationScheme as QuantizationMode
 from nncf.common.quantization.structs import QuantizerConfig
 from nncf.common.quantization.structs import UnifiedScaleType
 from nncf.common.scopes import should_consider_scope

nncf/common/quantization/quantizer_propagation/solver.py

@@ -40,7 +40,7 @@
 from nncf.common.quantization.quantizer_setup import SingleConfigQuantizerSetup
 from nncf.common.quantization.structs import QuantizableWeightedLayerNode
 from nncf.common.quantization.structs import QuantizationConstraints
-from nncf.common.quantization.structs import QuantizationMode
+from nncf.common.quantization.structs import QuantizationScheme as QuantizationMode
 from nncf.common.quantization.structs import QuantizerConfig
 from nncf.common.quantization.structs import QuantizerGroup
 from nncf.common.quantization.structs import UnifiedScaleType

nncf/common/quantization/quantizer_setup.py

@@ -18,7 +18,7 @@
 from nncf.common.graph import NNCFNodeName
 from nncf.common.logging import nncf_logger
 from nncf.common.quantization.structs import NonWeightQuantizerId
-from nncf.common.quantization.structs import QuantizationMode
+from nncf.common.quantization.structs import QuantizationScheme as QuantizationMode
 from nncf.common.quantization.structs import QuantizerConfig
 from nncf.common.quantization.structs import UnifiedScaleType
 from nncf.common.quantization.structs import WeightQuantizerId

nncf/common/quantization/structs.py

@@ -22,9 +22,9 @@
 
 
 @api()
-class QuantizationMode:
+class QuantizationScheme:
     """
-    Basic enumeration for quantization mode specification.
+    Basic enumeration for quantization scheme specification.
 
     :param SYMMETRIC:
     :param ASYMMETRIC:
@@ -43,7 +43,7 @@ class QuantizerConfig:
     def __init__(
         self,
         num_bits: int = QUANTIZATION_BITS,
-        mode: QuantizationMode = QuantizationMode.SYMMETRIC,
+        mode: QuantizationScheme = QuantizationScheme.SYMMETRIC,
         signedness_to_force: Optional[bool] = None,
         per_channel: bool = QUANTIZATION_PER_CHANNEL,
     ):
@@ -66,7 +66,7 @@ def __eq__(self, other):
     def __str__(self):
         return "B:{bits} M:{mode} SGN:{signedness} PC:{per_channel}".format(
             bits=self.num_bits,
-            mode="S" if self.mode == QuantizationMode.SYMMETRIC else "A",
+            mode="S" if self.mode == QuantizationScheme.SYMMETRIC else "A",
             signedness="ANY" if self.signedness_to_force is None else ("S" if self.signedness_to_force else "U"),
             per_channel="Y" if self.per_channel else "N",
         )
@@ -86,7 +86,7 @@ def is_valid_requantization_for(self, other: "QuantizerConfig") -> bool:
         """
         fail_conditions = [
             self.num_bits > other.num_bits,
-            self.mode is QuantizationMode.ASYMMETRIC and other.mode is QuantizationMode.SYMMETRIC,
+            self.mode is QuantizationScheme.ASYMMETRIC and other.mode is QuantizationScheme.SYMMETRIC,
             self.signedness_to_force is None and other.signedness_to_force is not None,
             self.signedness_to_force is True and other.signedness_to_force is False,
         ]
@@ -153,7 +153,7 @@ class QuantizerSpec:
     """
 
     def __init__(
-        self, num_bits: int, mode: QuantizationMode, signedness_to_force: bool, narrow_range: bool, half_range: bool
+        self, num_bits: int, mode: QuantizationScheme, signedness_to_force: bool, narrow_range: bool, half_range: bool
     ):
         """
         :param num_bits: Bitwidth of the quantization.
@@ -334,5 +334,5 @@ class QuantizationPreset(Enum):
 
     def get_params_configured_by_preset(self, quant_group: QuantizerGroup) -> Dict:
         if quant_group == QuantizerGroup.ACTIVATIONS and self == QuantizationPreset.MIXED:
-            return {"mode": QuantizationMode.ASYMMETRIC}
-        return {"mode": QuantizationMode.SYMMETRIC}
+            return {"mode": QuantizationScheme.ASYMMETRIC}
+        return {"mode": QuantizationScheme.SYMMETRIC}

nncf/experimental/tensor/functions.py

@@ -10,13 +10,15 @@
 # limitations under the License.
 
 import functools
-from typing import Callable, List, Optional, Tuple, Union
+from typing import Callable, List, Optional, Tuple, TypeVar, Union
 
 from nncf.experimental.tensor.enums import TensorDataType
 from nncf.experimental.tensor.enums import TensorDeviceType
 from nncf.experimental.tensor.tensor import Tensor
 from nncf.experimental.tensor.tensor import unwrap_tensor_data
 
+TypeInfo = TypeVar("TypeInfo")
+
 
 def _tensor_guard(func: callable):
     """
@@ -428,6 +430,18 @@ def _binary_reverse_op_nowarn(a: Tensor, b: Union[Tensor, float], operator_fn: C
     return Tensor(_binary_reverse_op_nowarn(a.data, unwrap_tensor_data(b), operator_fn))
 
 
+@functools.singledispatch
+@_tensor_guard
+def finfo(a: Tensor) -> TypeInfo:
+    """
+    Returns machine limits for tensor type.
+
+    :param a: Tensor.
+    :return: TypeInfo.
+    """
+    return finfo(a.data)
+
+
 def _dispatch_list(fn: "functools._SingleDispatchCallable", tensor_list: List[Tensor], *args, **kwargs):
     """
     Dispatches the function to the type of the wrapped data of the first element in tensor_list.

nncf/experimental/tensor/numpy_functions.py

@@ -206,3 +206,8 @@ def _(
     # Run operator with disabled warning
     with np.errstate(invalid="ignore", divide="ignore"):
         return operator_fn(b, a)
+
+
+@_register_numpy_types(fns.finfo)
+def _(a: np.ndarray) -> np.finfo:
+    return np.finfo(a.dtype)

nncf/experimental/tensorflow/quantization/quantizers.py

@@ -13,7 +13,7 @@
 
 import tensorflow as tf
 
-from nncf.common.quantization.structs import QuantizationMode
+from nncf.common.quantization.structs import QuantizationScheme as QuantizationMode
 from nncf.common.utils.registry import Registry
 from nncf.tensorflow.layers.operation import InputType
 from nncf.tensorflow.quantization.quantizers import AsymmetricQuantizer

nncf/onnx/quantization/quantize_model.py

@@ -9,7 +9,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from typing import Optional, Union
+from typing import Optional
 
 import onnx
 
@@ -18,8 +18,10 @@
 from nncf.data import Dataset
 from nncf.onnx.graph.nncf_graph_builder import GraphConverter
 from nncf.parameters import ModelType
+from nncf.parameters import QuantizationMode
 from nncf.parameters import TargetDevice
 from nncf.quantization.advanced_parameters import AdvancedQuantizationParameters
+from nncf.quantization.advanced_parameters import QuantizationParameters
 from nncf.quantization.algorithms.post_training.algorithm import PostTrainingQuantization
 from nncf.quantization.telemetry_extractors import CompressionStartedWithQuantizeApi
 from nncf.scopes import IgnoredScope
@@ -31,10 +33,11 @@
 def quantize_impl(
     model: onnx.ModelProto,
     calibration_dataset: Dataset,
-    preset: Union[QuantizationPreset, None],
-    target_device: TargetDevice,
-    subset_size: int,
-    fast_bias_correction: bool,
+    mode: Optional[QuantizationMode] = None,
+    preset: Optional[QuantizationPreset] = None,
+    target_device: TargetDevice = TargetDevice.ANY,
+    subset_size: int = 300,
+    fast_bias_correction: bool = True,
     model_type: Optional[ModelType] = None,
     ignored_scope: Optional[IgnoredScope] = None,
     advanced_parameters: Optional[AdvancedQuantizationParameters] = None,
@@ -44,6 +47,8 @@ def quantize_impl(
     """
     if target_device == TargetDevice.CPU_SPR:
         raise RuntimeError("target_device == CPU_SPR is not supported.")
+    if mode is not None:
+        raise ValueError(f"mode={mode} is not supported")
     if model.opset_import[0].version < 10:
         raise RuntimeError("ONNX models with opset version < 10 do not support quantization.")
     if model.opset_import[0].version < 13:
@@ -53,8 +58,8 @@ def quantize_impl(
         )
         if advanced_parameters is None:
             advanced_parameters = AdvancedQuantizationParameters()
-        advanced_parameters.weights_quantization_params.per_channel = False
-        advanced_parameters.activations_quantization_params.per_channel = False
+        advanced_parameters.weights_quantization_params = QuantizationParameters(per_channel=False)
+        advanced_parameters.activations_quantization_params = QuantizationParameters(per_channel=False)
 
     quantization_algorithm = PostTrainingQuantization(
         preset=preset,

nncf/openvino/graph/metatypes/groups.py

@@ -85,9 +85,7 @@
 ]
 
 
-FAKE_QUANTIZE_OPERATIONS = [
-    ov_metatypes.OVFakeQuantizeMetatype,
-]
+FAKE_QUANTIZE_OPERATIONS = [ov_metatypes.OVFakeQuantizeMetatype, ov_metatypes.OVFakeConvertMetatype]
 
 
 CONSTANT_OPERATIONS = [

nncf/openvino/graph/metatypes/openvino_metatypes.py

@@ -333,6 +333,12 @@ class OVFakeQuantizeMetatype(OVOpMetatype):
     op_names = ["FakeQuantize"]
 
 
+@OV_OPERATOR_METATYPES.register()
+class OVFakeConvertMetatype(OVOpMetatype):
+    name = "FakeConvertOp"
+    op_names = ["FakeConvert"]
+
+
 @OV_OPERATOR_METATYPES.register()
 class OVLessMetatype(OVOpMetatype):
     name = "LessOp"
@@ -713,13 +719,13 @@ def get_operation_const_op(operation: ov.Node, const_port_id: int) -> Optional[o
     # There are several cases here
     # (Constant) -> (Operation)
     # (Constant) -> (Convert) -> (Operation)
-    # (Constant) -> (Convert) -> (FakeQuantize) -> (Operation)
-    # (Constant) -> (Convert) -> (FakeQuantize) -> (Reshape) -> (Operation)
+    # (Constant) -> (Convert) -> (FakeQuantize, FakeConvert) -> (Operation)
+    # (Constant) -> (Convert) -> (FakeQuantize, FakeConvert) -> (Reshape) -> (Operation)
     #  and etc. We need properly find the constant node. So we start with
     # `node` and traverse up until the constant node is not found.
     queue = deque([node])
     constant_node = None
-    allowed_propagation_types_list = ["Convert", "FakeQuantize", "Reshape"]
+    allowed_propagation_types_list = ["Convert", "FakeQuantize", "FakeConvert", "Reshape"]
 
     while len(queue) != 0:
         curr_node = queue.popleft()