mirror of
https://github.com/PaddlePaddle/FastDeploy.git
synced 2025-10-03 15:56:49 +08:00
106 lines
3.7 KiB
Python
106 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.
|
|
|
|
import unittest
|
|
|
|
import numpy as np
|
|
import paddle
|
|
import paddle.nn.functional as F
|
|
|
|
from fastdeploy.model_executor.ops.gpu import group_swiglu_with_masked
|
|
|
|
paddle.seed(2024)
|
|
|
|
|
|
def group_swiglu_with_masked_paddle(fc1_out_tensor, token_nums_per_expert):
|
|
group_num, group_size, hidden_dim_x2 = fc1_out_tensor.shape
|
|
|
|
if token_nums_per_expert.dtype not in [paddle.int32, paddle.int64]:
|
|
raise ValueError(f"token_nums_per_expert must be int32 or int64, but receive {token_nums_per_expert.dtype}")
|
|
|
|
gate, up = paddle.chunk(fc1_out_tensor, chunks=2, axis=-1)
|
|
|
|
act_out = (F.silu(gate.to(paddle.float32)) * up.to(paddle.float32)).to(fc1_out_tensor.dtype)
|
|
|
|
# [0, 1, 2, ..., group_size-1]
|
|
range_tensor = paddle.arange(group_size, dtype=token_nums_per_expert.dtype)
|
|
|
|
mask = range_tensor < token_nums_per_expert.unsqueeze(1)
|
|
|
|
mask = mask.unsqueeze(-1)
|
|
|
|
output_tensor = act_out * mask.astype(act_out.dtype)
|
|
|
|
return output_tensor
|
|
|
|
|
|
class TestGroupSwigluWithMasked(unittest.TestCase):
|
|
def get_input(self):
|
|
self.token_nums_tensor = paddle.to_tensor([5, 8, 0, 3], dtype=self.token_nums_per_expert_dtype)
|
|
self.input_tensor = paddle.randn([self.group_num, self.group_size, self.hidden_dim * 2], dtype="bfloat16")
|
|
|
|
def setUp(self) -> None:
|
|
self.group_num = 4
|
|
self.group_size = 8
|
|
self.hidden_dim = 16 # fc1_out_tensor.shape()[2] / 2
|
|
self.input_dtype = paddle.bfloat16
|
|
self.token_nums_per_expert_dtype = paddle.int64
|
|
self.get_input()
|
|
|
|
def test_group_swiglu_with_masked(self):
|
|
paddle_output = group_swiglu_with_masked_paddle(self.input_tensor, self.token_nums_tensor)
|
|
output = group_swiglu_with_masked(self.input_tensor, self.token_nums_tensor)
|
|
|
|
valid_token_mask = paddle.arange(
|
|
self.group_size, dtype=self.token_nums_per_expert_dtype
|
|
) < self.token_nums_tensor.unsqueeze(1)
|
|
|
|
# Note(ooooo): Because GetEmptyTensor will random.
|
|
np.testing.assert_allclose(
|
|
paddle_output[valid_token_mask].astype("float32").numpy(),
|
|
output[valid_token_mask].astype("float32").numpy(),
|
|
)
|
|
|
|
|
|
class TestGroupSwigluWithMaskedCase1(TestGroupSwigluWithMasked):
|
|
def setUp(self) -> None:
|
|
self.group_num = 4
|
|
self.group_size = 8
|
|
self.hidden_dim = 16 # fc1_out_tensor.shape()[2] / 2
|
|
self.input_dtype = paddle.bfloat16
|
|
self.token_nums_per_expert_dtype = paddle.int32
|
|
self.get_input()
|
|
|
|
|
|
class TestGroupSwigluWithMaskedCase2(TestGroupSwigluWithMasked):
|
|
def setUp(self) -> None:
|
|
self.group_num = 4
|
|
self.group_size = 8
|
|
self.hidden_dim = 16 # fc1_out_tensor.shape()[2] / 2
|
|
self.input_dtype = paddle.bfloat16
|
|
self.token_nums_per_expert_dtype = paddle.int32
|
|
self.get_input()
|
|
|
|
def get_input(self):
|
|
self.token_nums_tensor = paddle.randint(
|
|
0, self.group_size + 1, shape=[self.group_num], dtype=self.token_nums_per_expert_dtype
|
|
)
|
|
self.input_tensor = paddle.randn(
|
|
[self.group_num, self.group_size, self.hidden_dim * 2], dtype=self.input_dtype
|
|
)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|