wehub-resource-sync
386 lines
14 KiB
Python
386 lines
14 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.
|
|
from __future__ import annotations
|
|
|
|
import unittest
|
|
from dataclasses import dataclass
|
|
|
|
import numpy as np
|
|
from dist_amp_base import (
|
|
create_optimizer,
|
|
)
|
|
|
|
import paddle
|
|
from paddle import nn
|
|
from paddle.distributed import fleet
|
|
from paddle.distributed.fleet.meta_parallel import (
|
|
LayerDesc,
|
|
LayerSpec,
|
|
NoPipelineParallel,
|
|
PipelineLayer,
|
|
build_spec_layer,
|
|
)
|
|
from paddle.distributed.fleet.meta_parallel.pipeline_parallel import (
|
|
PipelineDatasetPreprocessor,
|
|
)
|
|
from paddle.nn import Layer
|
|
from paddle.nn.layer import Identity
|
|
|
|
|
|
class ReshapeHelp(Layer):
|
|
def __init__(self, shape):
|
|
super().__init__()
|
|
self.shape = shape
|
|
|
|
def forward(self, x):
|
|
return x.reshape(shape=self.shape)
|
|
|
|
|
|
@dataclass
|
|
class AlexNetLayerSpec:
|
|
features: list[LayerSpec] | list[Identity]
|
|
reshape_layer: LayerSpec | type = Identity
|
|
classifier: LayerSpec | type = Identity
|
|
|
|
|
|
class AlexNet(PipelineLayer):
|
|
def __init__(self, sublayers_spec: AlexNetLayerSpec, **kwargs):
|
|
self.layers = AlexNet.get_layer_desc_list(sublayers_spec)
|
|
|
|
super().__init__(layers=self.layers, **kwargs)
|
|
|
|
@staticmethod
|
|
def get_layer_desc_list(spec: AlexNetLayerSpec):
|
|
layers = []
|
|
for features_spec in spec.features:
|
|
layers.append(LayerDesc(features_spec))
|
|
layers.append(LayerDesc(spec.reshape_layer))
|
|
layers.append(LayerDesc(spec.classifier))
|
|
return layers
|
|
|
|
|
|
def get_alex_spec(num_classes=10):
|
|
spec = LayerSpec(
|
|
layer=AlexNet,
|
|
sublayers_spec=AlexNetLayerSpec(
|
|
features=[
|
|
LayerSpec(
|
|
layer=nn.Conv2D,
|
|
extra_kwargs={
|
|
"in_channels": 3,
|
|
"out_channels": 3,
|
|
"kernel_size": 11,
|
|
"stride": 4,
|
|
"padding": 5,
|
|
},
|
|
),
|
|
LayerSpec(
|
|
layer=nn.ReLU,
|
|
),
|
|
LayerSpec(
|
|
layer=nn.MaxPool2D,
|
|
extra_kwargs={"kernel_size": 2, "stride": 2},
|
|
),
|
|
LayerSpec(
|
|
layer=nn.Conv2D,
|
|
extra_kwargs={
|
|
"in_channels": 3,
|
|
"out_channels": 3,
|
|
"kernel_size": 5,
|
|
"padding": 2,
|
|
},
|
|
),
|
|
LayerSpec(
|
|
layer=nn.ReLU,
|
|
),
|
|
LayerSpec(
|
|
layer=nn.MaxPool2D,
|
|
extra_kwargs={"kernel_size": 2, "stride": 2},
|
|
),
|
|
LayerSpec(
|
|
layer=nn.Conv2D,
|
|
extra_kwargs={
|
|
"in_channels": 3,
|
|
"out_channels": 3,
|
|
"kernel_size": 3,
|
|
"padding": 1,
|
|
},
|
|
),
|
|
LayerSpec(
|
|
layer=nn.ReLU,
|
|
),
|
|
LayerSpec(
|
|
layer=nn.Conv2D,
|
|
extra_kwargs={
|
|
"in_channels": 3,
|
|
"out_channels": 3,
|
|
"kernel_size": 3,
|
|
"padding": 1,
|
|
},
|
|
),
|
|
LayerSpec(
|
|
layer=nn.ReLU,
|
|
),
|
|
LayerSpec(
|
|
layer=nn.Conv2D,
|
|
extra_kwargs={
|
|
"in_channels": 3,
|
|
"out_channels": 3,
|
|
"kernel_size": 3,
|
|
"padding": 1,
|
|
},
|
|
),
|
|
LayerSpec(
|
|
layer=nn.ReLU,
|
|
),
|
|
LayerSpec(
|
|
layer=nn.MaxPool2D,
|
|
extra_kwargs={"kernel_size": 2, "stride": 2},
|
|
),
|
|
],
|
|
reshape_layer=LayerSpec(
|
|
layer=ReshapeHelp, extra_kwargs={"shape": [-1, 256]}
|
|
),
|
|
classifier=LayerSpec(
|
|
layer=nn.Linear,
|
|
extra_kwargs={"in_features": 256, "out_features": num_classes},
|
|
),
|
|
),
|
|
extra_kwargs={
|
|
"loss_fn": nn.CrossEntropyLoss(),
|
|
},
|
|
)
|
|
return spec
|
|
|
|
|
|
class TestPipeLayerAPI(unittest.TestCase):
|
|
def setUp(self):
|
|
strategy = fleet.DistributedStrategy()
|
|
self.pipeline_parallel_size = 2
|
|
strategy.hybrid_configs = {
|
|
"dp_degree": 1,
|
|
"mp_degree": 1,
|
|
"pp_degree": self.pipeline_parallel_size,
|
|
}
|
|
batch_size = 8
|
|
micro_batch_size = 2
|
|
strategy.pipeline_configs = {
|
|
"accumulate_steps": batch_size // micro_batch_size,
|
|
"micro_batch_size": micro_batch_size,
|
|
}
|
|
self.strategy = strategy
|
|
fleet.init(is_collective=True, strategy=strategy)
|
|
self.hcg = fleet.get_hybrid_communicate_group()
|
|
|
|
def test_pipelayer_desc(self):
|
|
alex_desc = get_alex_spec()
|
|
pipe_model = build_spec_layer(
|
|
alex_desc, num_stages=self.pipeline_parallel_size
|
|
)
|
|
np.testing.assert_array_equal(len(pipe_model.parameters()), 6)
|
|
|
|
def test_pipelayer_desc_single(self):
|
|
alex_desc = get_alex_spec()
|
|
pipe_model = build_spec_layer(alex_desc, num_stages=1)
|
|
np.testing.assert_array_equal(len(pipe_model.parameters()), 12)
|
|
pipe_model = NoPipelineParallel(pipe_model, self.strategy, self.hcg)
|
|
input = paddle.randn([256, 3, 224, 224])
|
|
label = paddle.randint(0, 10, [147, 1])
|
|
|
|
# Test with list data
|
|
data = [[input, input, input, input], [label, label, label, label]]
|
|
optimizer = create_optimizer(
|
|
model=pipe_model, use_pure_bf16=True, use_main_grad=True
|
|
)
|
|
base_lr = 0.1
|
|
lr_scheduler = paddle.optimizer.lr.CosineAnnealingDecay(
|
|
learning_rate=base_lr, T_max=1
|
|
)
|
|
pipe_model.train_batch(data, optimizer, lr_scheduler)
|
|
pipe_model.eval_batch(data, optimizer)
|
|
|
|
pipe_model.train_batch(data, optimizer)
|
|
pipe_model.eval_batch(data, optimizer)
|
|
pipe_model.is_pipeline_last_stage()
|
|
|
|
pipe_model.train_batch(data, optimizer, return_micro_batch_loss=True)
|
|
|
|
scaler = paddle.amp.GradScaler(init_loss_scaling=4096)
|
|
scaler = fleet.distributed_scaler(scaler)
|
|
pipe_model.train_batch(data, optimizer, scaler=scaler)
|
|
|
|
def test_pipelayer_segment_method_list(self):
|
|
alex_desc = get_alex_spec()
|
|
pipe_model = build_spec_layer(
|
|
alex_desc, num_stages=self.pipeline_parallel_size, seg_method=[0, 4]
|
|
)
|
|
stage_id = self.hcg.get_stage_id()
|
|
if stage_id == 0:
|
|
np.testing.assert_array_equal(len(pipe_model.parameters()), 4)
|
|
elif stage_id == 1:
|
|
np.testing.assert_array_equal(len(pipe_model.parameters()), 8)
|
|
|
|
def test_pipelayer_segment_method_spec(self):
|
|
alex_desc = get_alex_spec()
|
|
pipe_model = build_spec_layer(
|
|
alex_desc,
|
|
num_stages=self.pipeline_parallel_size,
|
|
seg_method="layer:Conv2D|MaxPool2D",
|
|
)
|
|
stage_id = self.hcg.get_stage_id()
|
|
if stage_id == 0:
|
|
np.testing.assert_array_equal(len(pipe_model.parameters()), 4)
|
|
elif stage_id == 1:
|
|
np.testing.assert_array_equal(len(pipe_model.parameters()), 8)
|
|
|
|
def test_pipelayer_segment_method_vpp(self):
|
|
alex_desc = get_alex_spec()
|
|
pipe_model = build_spec_layer(
|
|
alex_desc,
|
|
num_stages=self.pipeline_parallel_size,
|
|
seg_method="layer:Conv2D|MaxPool2D",
|
|
num_virtual_pipeline_stages=2,
|
|
)
|
|
stage_id = self.hcg.get_stage_id()
|
|
if stage_id == 0:
|
|
np.testing.assert_array_equal(len(pipe_model.parameters()), 6)
|
|
elif stage_id == 1:
|
|
np.testing.assert_array_equal(len(pipe_model.parameters()), 6)
|
|
|
|
def test_check_micro_batch_data_valid_with_tuple(self):
|
|
"""Test _check_micro_batch_data_valid with tuple data."""
|
|
|
|
alex_desc = get_alex_spec()
|
|
pipe_model = build_spec_layer(alex_desc, num_stages=1)
|
|
pipe_model = NoPipelineParallel(pipe_model, self.strategy, self.hcg)
|
|
|
|
# Test with tuple data
|
|
tensor1 = paddle.randn([2, 3])
|
|
tensor2 = paddle.randn([2, 3])
|
|
tuple_data = (tensor1, tensor2)
|
|
|
|
# This should not raise any exception
|
|
pipe_model._check_micro_batch_data_valid(tuple_data)
|
|
|
|
def test_check_micro_batch_data_valid_with_dict(self):
|
|
"""Test _check_micro_batch_data_valid with dict data."""
|
|
|
|
alex_desc = get_alex_spec()
|
|
pipe_model = build_spec_layer(alex_desc, num_stages=1)
|
|
pipe_model = NoPipelineParallel(pipe_model, self.strategy, self.hcg)
|
|
|
|
# Test with dict data
|
|
dict_data = {
|
|
"input": paddle.randn([2, 3]),
|
|
"label": paddle.randn([2, 1]),
|
|
}
|
|
|
|
# This should not raise any exception
|
|
pipe_model._check_micro_batch_data_valid(dict_data)
|
|
|
|
def test_eval_batch_with_pipeline_dataset_preprocessor(self):
|
|
"""Test eval_batch with wrapper PipelineDatasetPreprocessor."""
|
|
|
|
alex_desc = get_alex_spec()
|
|
pipe_model = build_spec_layer(alex_desc, num_stages=1)
|
|
pipe_model = NoPipelineParallel(pipe_model, self.strategy, self.hcg)
|
|
|
|
input = paddle.randn([256, 3, 224, 224])
|
|
label = paddle.randint(0, 10, [147, 1])
|
|
|
|
# Test with list data
|
|
data = [[input, input, input, input], [label, label, label, label]]
|
|
|
|
# Test with PipelineDatasetPreprocessor wrapper
|
|
def data_generator():
|
|
return data
|
|
|
|
preprocessed_data = PipelineDatasetPreprocessor(data_generator)
|
|
# This should work - calling preprocessed_data() should return the data
|
|
result = preprocessed_data()
|
|
self.assertEqual(result, data)
|
|
|
|
# Test eval_batch with this preprocessed data
|
|
pipe_model.eval_batch(preprocessed_data)
|
|
|
|
def test_eval_batch_with_non_tuple_list_data(self):
|
|
"""Test eval_batch with non-tuple/list data (iterable like generator)."""
|
|
from paddle.distributed.fleet.meta_parallel.pipeline_parallel import (
|
|
NoPipelineParallel,
|
|
)
|
|
|
|
alex_desc = get_alex_spec()
|
|
pipe_model = build_spec_layer(alex_desc, num_stages=1)
|
|
pipe_model = NoPipelineParallel(pipe_model, self.strategy, self.hcg)
|
|
|
|
# Test with a generator (not tuple or list) - this covers the branch
|
|
# where data is not tuple/list but is an iterable
|
|
input = paddle.randn([256, 3, 224, 224])
|
|
label = paddle.randint(0, 10, [147, 1])
|
|
|
|
# Create a generator that yields (input, label) pairs
|
|
def data_generator():
|
|
for _ in range(pipe_model.accumulate_steps):
|
|
yield (input, label)
|
|
|
|
# This should work - generator is not tuple/list, so it goes to micro_dataset directly
|
|
# Note: This test validates the code path, actual behavior depends on implementation
|
|
try:
|
|
pipe_model.eval_batch(data_generator)
|
|
except (TypeError, StopIteration):
|
|
# The generator gets exhausted or other expected errors
|
|
pass
|
|
|
|
def test_eval_batch_return_host_tensor(self):
|
|
"""Test eval_batch with return_host_tensor=True, covering _offload_tensors."""
|
|
alex_desc = get_alex_spec()
|
|
pipe_model = build_spec_layer(alex_desc, num_stages=1)
|
|
pipe_model = NoPipelineParallel(pipe_model, self.strategy, self.hcg)
|
|
|
|
input = paddle.randn([256, 3, 224, 224])
|
|
label = paddle.randint(0, 10, [147, 1])
|
|
data = [[input, input, input, input], [label, label, label, label]]
|
|
|
|
# compute_loss=False, return_host_tensor=True → triggers _offload_tensors
|
|
# with a single Tensor output (lines 680, 710-718)
|
|
result = pipe_model.eval_batch(
|
|
data, compute_loss=False, return_host_tensor=True
|
|
)
|
|
self.assertIsInstance(result, list)
|
|
self.assertEqual(len(result), pipe_model.accumulate_steps)
|
|
|
|
def test_offload_tensors_branches(self):
|
|
"""Directly test _offload_tensors covering all branches (lines 702-718)."""
|
|
alex_desc = get_alex_spec()
|
|
pipe_model = build_spec_layer(alex_desc, num_stages=1)
|
|
pipe_model = NoPipelineParallel(pipe_model, self.strategy, self.hcg)
|
|
|
|
t = paddle.randn([4, 4])
|
|
|
|
# Branch: single Tensor (lines 710-718)
|
|
pipe_model._offload_tensors(t)
|
|
|
|
# Branch: single non-Tensor → early return (line 712)
|
|
pipe_model._offload_tensors("not_a_tensor")
|
|
|
|
# Branch: tuple with Tensor elements (lines 702-709)
|
|
pipe_model._offload_tensors((t, paddle.randn([2, 2])))
|
|
|
|
# Branch: tuple containing a non-Tensor element → continue (lines 703-705)
|
|
pipe_model._offload_tensors((t, "not_a_tensor"))
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|