wehub-resource-sync
152 lines
5.2 KiB
Python
152 lines
5.2 KiB
Python
# Copyright (c) 2023 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
|
|
|
|
from paddle.distributed.auto_parallel.static.dist_attribute import (
|
|
DistTensorSpec,
|
|
TensorDistAttr,
|
|
)
|
|
from paddle.distributed.fleet import auto
|
|
from paddle.framework import core
|
|
|
|
|
|
class TestFlashAttentionSPMDRule(unittest.TestCase):
|
|
"""
|
|
Unit tests for layer_norm spmd rule.
|
|
"""
|
|
|
|
def setUp(self):
|
|
self.rule = core.get_phi_spmd_rule("flash_attention")
|
|
process_mesh = auto.ProcessMesh(mesh=[[0, 1, 2], [3, 4, 5]])
|
|
self.process_mesh = process_mesh
|
|
|
|
q_tensor_dist_attr = TensorDistAttr()
|
|
q_tensor_dist_attr.process_mesh = process_mesh
|
|
q_tensor_dist_attr.dims_mapping = [0, -1, 1, -1]
|
|
q_shape = [2, 512, 64, 1024]
|
|
q_spec = DistTensorSpec(q_shape, q_tensor_dist_attr)
|
|
self.q_spec = q_spec
|
|
|
|
k_tensor_dist_attr = TensorDistAttr()
|
|
k_tensor_dist_attr.process_mesh = process_mesh
|
|
k_tensor_dist_attr.dims_mapping = [0, -1, -1, -1]
|
|
k_shape = [2, 1024, 64, 1024]
|
|
k_spec = DistTensorSpec(k_shape, k_tensor_dist_attr)
|
|
self.k_spec = k_spec
|
|
|
|
v_tensor_dist_attr = TensorDistAttr()
|
|
v_tensor_dist_attr.process_mesh = process_mesh
|
|
v_tensor_dist_attr.dims_mapping = [0, -1, -1, -1]
|
|
v_shape = [2, 1024, 64, 512]
|
|
v_spec = DistTensorSpec(v_shape, v_tensor_dist_attr)
|
|
self.v_spec = v_spec
|
|
|
|
out_tensor_dist_attr = TensorDistAttr()
|
|
out_tensor_dist_attr.process_mesh = process_mesh
|
|
out_tensor_dist_attr.dims_mapping = [0, -1, 1, -1]
|
|
out_shape = [2, 512, 64, 512]
|
|
out_spec = DistTensorSpec(out_shape, out_tensor_dist_attr)
|
|
self.out_spec = out_spec
|
|
|
|
softmax_lse_dist_attr = TensorDistAttr()
|
|
softmax_lse_dist_attr.process_mesh = process_mesh
|
|
softmax_lse_dist_attr.dims_mapping = [0, 1, -1]
|
|
softmax_lse_shape = [2, 64, 512]
|
|
softmax_lse_spec = DistTensorSpec(
|
|
softmax_lse_shape, softmax_lse_dist_attr
|
|
)
|
|
self.softmax_lse_spec = softmax_lse_spec
|
|
|
|
def create_empty_tensor(self):
|
|
dist_attr = TensorDistAttr()
|
|
dist_attr.process_mesh = self.process_mesh
|
|
dist_attr.dims_mapping = []
|
|
shape = []
|
|
return DistTensorSpec(shape, dist_attr)
|
|
|
|
def test_infer_forward(self):
|
|
result_dist_attrs = self.rule.infer_forward(
|
|
self.q_spec,
|
|
self.k_spec,
|
|
self.v_spec,
|
|
self.create_empty_tensor(),
|
|
self.create_empty_tensor(),
|
|
0.0,
|
|
False,
|
|
False,
|
|
False,
|
|
)
|
|
inferred_input_dist_attrs = result_dist_attrs[0]
|
|
inferred_output_dist_attrs = result_dist_attrs[1]
|
|
|
|
self.assertEqual(len(result_dist_attrs), 2)
|
|
self.assertEqual(len(inferred_input_dist_attrs), 5)
|
|
self.assertEqual(len(inferred_output_dist_attrs), 4)
|
|
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[0].dims_mapping, [0, -1, 1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[1].dims_mapping, [0, -1, 1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[2].dims_mapping, [0, -1, 1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_output_dist_attrs[0].dims_mapping, [0, -1, 1, -1]
|
|
)
|
|
self.assertEqual(inferred_output_dist_attrs[2].dims_mapping, [0, 1, -1])
|
|
|
|
def test_infer_backward(self):
|
|
result_dist_attrs = self.rule.infer_backward(
|
|
self.q_spec,
|
|
self.k_spec,
|
|
self.v_spec,
|
|
self.create_empty_tensor(),
|
|
self.create_empty_tensor(),
|
|
self.out_spec,
|
|
self.create_empty_tensor(),
|
|
self.softmax_lse_spec,
|
|
self.create_empty_tensor(),
|
|
0.0,
|
|
False,
|
|
False,
|
|
False,
|
|
)
|
|
inferred_input_dist_attrs = result_dist_attrs[0]
|
|
inferred_output_dist_attrs = result_dist_attrs[1]
|
|
|
|
self.assertEqual(len(result_dist_attrs), 2)
|
|
self.assertEqual(len(inferred_input_dist_attrs), 5)
|
|
self.assertEqual(len(inferred_output_dist_attrs), 4)
|
|
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[0].dims_mapping, [0, -1, 1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[1].dims_mapping, [0, -1, 1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_input_dist_attrs[2].dims_mapping, [0, -1, 1, -1]
|
|
)
|
|
self.assertEqual(
|
|
inferred_output_dist_attrs[0].dims_mapping, [0, -1, 1, -1]
|
|
)
|
|
self.assertEqual(inferred_output_dist_attrs[2].dims_mapping, [0, 1, -1])
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|