mirror of
https://github.com/PaddlePaddle/FastDeploy.git
synced 2025-11-01 04:12:58 +08:00
94 lines
3.7 KiB
Python
94 lines
3.7 KiB
Python
"""
|
|
# Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved.
|
|
#
|
|
# 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.
|
|
|
|
!! This file will be deleted after the platform is fully functional
|
|
"""
|
|
|
|
from typing import Tuple
|
|
|
|
import numpy as np
|
|
import paddle
|
|
|
|
|
|
def xpu_clip_and_round(x: np.ndarray) -> np.ndarray:
|
|
"""
|
|
Clip and round the input array to the range [-127, 127] and convert to int8.
|
|
|
|
Args:
|
|
x (numpy.ndarray): The input array to be clipped and rounded.
|
|
|
|
Returns:
|
|
numpy.ndarray: The clipped and rounded array with dtype int8.
|
|
"""
|
|
return np.clip(np.around(x), -127, 127).astype("int8")
|
|
|
|
|
|
def xpu_quant_qkv_weight(
|
|
weight_np: np.ndarray,
|
|
) -> Tuple[paddle.Tensor, paddle.Tensor]:
|
|
"""
|
|
Quantize the query, key, and value weights for the Transformer model.
|
|
|
|
Args:
|
|
weight_np (numpy.ndarray): The original weights of query, key, and value in numpy format.
|
|
It should be a 2D or higher dimensional tensor, where the last dimension represents the
|
|
embedding dimension.
|
|
|
|
Returns:
|
|
tuple: A tuple containing:
|
|
quanted_weight (paddle.Tensor): The quantized weights in paddle tensor format,
|
|
with the same shape as the input weight_np.
|
|
weight_scales (paddle.Tensor): The scaling factors for each element in the last dimension
|
|
of the input, used to recover the original value range from the quantized weights.
|
|
"""
|
|
dim_embed = weight_np.shape[-1]
|
|
weight = np.reshape(weight_np, [-1, dim_embed])
|
|
max_value = np.max(np.abs(weight), axis=1).reshape(-1, 1)
|
|
quanted_weight = xpu_clip_and_round(weight / max_value * 127.0)
|
|
quanted_weight = np.reshape(quanted_weight, weight_np.shape)
|
|
quanted_weight = paddle.to_tensor(quanted_weight, place=paddle.CPUPlace())
|
|
weight_scales = (max_value / 127.0).astype(weight_np.dtype).reshape(-1)
|
|
weight_scales = paddle.to_tensor(weight_scales, place=paddle.CPUPlace())
|
|
weight_scales = paddle.cast(weight_scales, paddle.get_default_dtype())
|
|
return quanted_weight, weight_scales
|
|
|
|
|
|
def xpu_quant_weight(
|
|
weight_np: np.ndarray,
|
|
) -> Tuple[paddle.Tensor, paddle.Tensor]:
|
|
"""
|
|
Quantize the weight tensor for XPU devices.
|
|
|
|
Args:
|
|
weight_np (numpy.ndarray): The original weight tensor in numpy format,
|
|
expected to be a 2D array.
|
|
|
|
Returns:
|
|
tuple: A tuple containing two elements:
|
|
quanted_weight (paddle.Tensor): The quantized weight tensor,
|
|
converted to a Paddle Tensor on CPU.
|
|
weight_scales (paddle.Tensor): The corresponding scales for the quantized
|
|
weights, also converted to a Paddle Tensor on CPU and cast to the
|
|
default data type.
|
|
"""
|
|
weight = np.transpose(weight_np, [1, 0])
|
|
max_value = np.max(np.abs(weight), axis=1).reshape(-1, 1)
|
|
quanted_weight = xpu_clip_and_round(weight / max_value * 127.0)
|
|
quanted_weight = paddle.to_tensor(quanted_weight, place=paddle.CPUPlace())
|
|
weight_scales = (max_value / 127.0).astype(weight_np.dtype).reshape(-1)
|
|
weight_scales = paddle.to_tensor(weight_scales, place=paddle.CPUPlace())
|
|
weight_scales = paddle.cast(weight_scales, paddle.get_default_dtype())
|
|
return quanted_weight, weight_scales
|