wehub-resource-sync
96 lines
2.8 KiB
Python
96 lines
2.8 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
|
|
from paddle.incubate.nn.functional import fused_partial_rope
|
|
|
|
|
|
def fused_partial_rope_ref(x, cos, sin):
|
|
x_nope = x[..., : -cos.shape[-1]]
|
|
x_pe = x[..., -cos.shape[-1] :]
|
|
|
|
b, s, h, d = x_pe.shape # [bs, seq_len, num_heads, pe_head_dim]
|
|
x_pe = (
|
|
x_pe.reshape([b, s, h, d // 2, 2])
|
|
.transpose([0, 1, 2, 4, 3])
|
|
.reshape([b, s, h, d])
|
|
)
|
|
|
|
cos = cos[:, :s, :, :] # [1, seq_len, 1, pe_head_dim]
|
|
sin = sin[:, :s, :, :]
|
|
|
|
x1 = x_pe[..., : x_pe.shape[-1] // 2]
|
|
x2 = x_pe[..., x_pe.shape[-1] // 2 :]
|
|
x_pe_rotate_half = paddle.concat([-x2, x1], axis=-1)
|
|
|
|
x_pe = (x_pe * cos) + (x_pe_rotate_half * sin)
|
|
|
|
return paddle.concat([x_nope, x_pe], axis=-1)
|
|
|
|
|
|
class TestFusedPartialRoPEOp(unittest.TestCase):
|
|
def eval(self, batch_size, seq_len, num_heads, head_dim, pe_head_dim):
|
|
x = paddle.randn([batch_size, seq_len, num_heads, head_dim], 'bfloat16')
|
|
x.stop_gradient = False
|
|
x_ref = paddle.clone(x).detach()
|
|
x_ref.stop_gradient = False
|
|
|
|
cos = paddle.randn([1, seq_len, 1, pe_head_dim], 'bfloat16')
|
|
sin = paddle.randn_like(cos)
|
|
|
|
# Test forward
|
|
out = fused_partial_rope(x, cos, sin)
|
|
out_ref = fused_partial_rope_ref(x_ref, cos, sin)
|
|
|
|
np.testing.assert_allclose(
|
|
out.astype('float32'), out_ref.astype('float32')
|
|
)
|
|
|
|
# Test backward
|
|
out_grad = paddle.randn_like(out)
|
|
paddle.autograd.backward([out], [out_grad])
|
|
paddle.autograd.backward([out_ref], [out_grad])
|
|
|
|
np.testing.assert_allclose(
|
|
x.grad.astype('float32'), x_ref.grad.astype('float32')
|
|
)
|
|
|
|
def test_0_size_in_batch_size(self):
|
|
self.eval(0, 32, 64, 128, 64)
|
|
|
|
def test_0_size_in_seq_len(self):
|
|
self.eval(32, 0, 64, 128, 64)
|
|
|
|
def test_all_pe_head_dim(self):
|
|
self.eval(1, 8, 1, 128, 128)
|
|
|
|
def test_medium_1x_vec(self):
|
|
self.eval(1, 8, 16, 75, 50)
|
|
|
|
def test_medium_2x_vec(self):
|
|
self.eval(4, 1, 16, 200, 100)
|
|
|
|
def test_medium_4x_vec(self):
|
|
self.eval(2, 4, 8, 192, 64)
|
|
|
|
def test_large(self):
|
|
self.eval(1, 2, 16, 1024, 384)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|