wehub-resource-sync
235 lines
10 KiB
Python
235 lines
10 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 copy
|
|
import math
|
|
import os
|
|
import re
|
|
import shutil
|
|
import tempfile
|
|
import unittest
|
|
from tempfile import TemporaryDirectory
|
|
|
|
import numpy as np
|
|
import paddle
|
|
from parameterized import parameterized
|
|
|
|
from paddlenlp.peft.lora import LoRAConfig, LoRALinear, LoRAModel
|
|
from paddlenlp.transformers import AutoModel, BertModel
|
|
from paddlenlp.utils.optimizer import AdamWLoRAPro
|
|
|
|
|
|
class TestLoRAProLayer(unittest.TestCase):
|
|
def test_r_raise_exception(self):
|
|
with self.assertRaises(ValueError):
|
|
LoRALinear(in_features=16, out_features=8, r=0, lora_dropout=0.1, lora_alpha=8, lorapro=True)
|
|
|
|
def test_forward(self):
|
|
lorapro_layer = LoRALinear(in_features=16, out_features=8, r=4, lora_dropout=0.1, lora_alpha=8, lorapro=True)
|
|
x = paddle.randn([2, 4, 16], "float32")
|
|
output = lorapro_layer(x)
|
|
self.assertFalse(lorapro_layer.lora_A.stop_gradient)
|
|
self.assertFalse(lorapro_layer.lora_B.stop_gradient)
|
|
self.assertTrue(lorapro_layer.weight.stop_gradient)
|
|
self.assertFalse(lorapro_layer.bias.stop_gradient)
|
|
self.assertEqual(output.shape, [2, 4, 8])
|
|
|
|
def test_train_eval(self):
|
|
x = paddle.randn([2, 4, 16], "float32")
|
|
lorapro_layer = LoRALinear(in_features=16, out_features=8, r=4, lorapro=True)
|
|
lorapro_layer.train()
|
|
train_result = lorapro_layer(x)
|
|
train_weight = copy.deepcopy(lorapro_layer.weight) # deep copy since this is a pointer
|
|
lorapro_layer.eval()
|
|
eval_result = lorapro_layer(x)
|
|
eval_weight = lorapro_layer.weight
|
|
self.assertTrue(paddle.allclose(train_result, eval_result))
|
|
self.assertTrue(paddle.allclose(train_weight, eval_weight))
|
|
|
|
def test_save_load(self):
|
|
with TemporaryDirectory() as tempdir:
|
|
lorapro_layer = LoRALinear(in_features=16, out_features=8, r=4, lorapro=True)
|
|
weights_path = os.path.join(tempdir, "model.pdparams")
|
|
paddle.save(lorapro_layer.state_dict(), weights_path)
|
|
new_lorapro_layer = LoRALinear(in_features=16, out_features=8, r=4, lorapro=True)
|
|
state_dict = paddle.load(weights_path)
|
|
new_lorapro_layer.set_dict(state_dict)
|
|
x = paddle.randn([2, 4, 16], "float32")
|
|
self.assertTrue(paddle.allclose(new_lorapro_layer(x), lorapro_layer(x)))
|
|
|
|
def test_load_regular_linear(self):
|
|
with TemporaryDirectory() as tempdir:
|
|
regular_linear = paddle.nn.Linear(in_features=16, out_features=8)
|
|
weights_path = os.path.join(tempdir, "model.pdparams")
|
|
paddle.save(regular_linear.state_dict(), weights_path)
|
|
state_dict = paddle.load(weights_path)
|
|
# should be identical to regular linear
|
|
lorapro_layer_r8 = LoRALinear(in_features=16, out_features=8, r=8, lorapro=True)
|
|
lorapro_layer_r4 = LoRALinear(in_features=16, out_features=8, r=4, lorapro=True)
|
|
lorapro_layer_r8.set_dict(state_dict)
|
|
lorapro_layer_r4.set_dict(state_dict)
|
|
x = paddle.randn([2, 4, 16], "float32")
|
|
self.assertTrue(paddle.allclose(lorapro_layer_r8(x), regular_linear(x)))
|
|
self.assertTrue(paddle.allclose(lorapro_layer_r4(x), regular_linear(x)))
|
|
|
|
|
|
class TestLoRAProModel(unittest.TestCase):
|
|
def setUp(self):
|
|
self.output_dir = tempfile.mkdtemp()
|
|
|
|
def tearDown(self):
|
|
shutil.rmtree(self.output_dir)
|
|
|
|
def test_lorapro_model_restore(self):
|
|
lorapro_config = LoRAConfig(
|
|
target_modules=[".*q_proj.*", ".*v_proj.*"],
|
|
r=4,
|
|
lora_alpha=8,
|
|
enable_lora_list=[None, [True, False]],
|
|
head_dim=2,
|
|
lorapro=True,
|
|
)
|
|
model = AutoModel.from_pretrained("__internal_testing__/tiny-random-bert")
|
|
input_ids = paddle.to_tensor(np.random.randint(100, 200, [1, 20]))
|
|
model.eval()
|
|
original_results_1 = model(input_ids)
|
|
lorapro_model = LoRAModel(model, lorapro_config)
|
|
restored_model = lorapro_model.restore_original_model()
|
|
restored_model.eval()
|
|
original_results_2 = restored_model(input_ids)
|
|
self.assertIsNotNone(original_results_1)
|
|
self.assertIsNotNone(original_results_2)
|
|
self.assertIsInstance(restored_model, BertModel)
|
|
self.assertTrue(paddle.allclose(original_results_1[0], original_results_2[0]))
|
|
|
|
@parameterized.expand([(None,), ("all",), ("lora",)])
|
|
def test_lorapro_model_constructor(self, bias):
|
|
lorapro_config = LoRAConfig(
|
|
target_modules=[".*q_proj.*", ".*v_proj.*"],
|
|
r=4,
|
|
lora_alpha=8,
|
|
enable_lora_list=[None, [True, False]],
|
|
trainable_bias=bias,
|
|
head_dim=2,
|
|
lorapro=True,
|
|
)
|
|
# turn off plm dropout for to test train vs test
|
|
model = AutoModel.from_pretrained(
|
|
"__internal_testing__/tiny-random-bert", hidden_dropout_prob=0, attention_probs_dropout_prob=0
|
|
)
|
|
lorapro_model = LoRAModel(model, lorapro_config)
|
|
lorapro_model.mark_only_lora_as_trainable()
|
|
for name, weight in lorapro_model.state_dict().items():
|
|
if any([re.fullmatch(target_module, name) for target_module in lorapro_config.target_modules]):
|
|
if "lora" in name:
|
|
self.assertFalse(weight.stop_gradient)
|
|
elif "bias" in name and bias in ["lora", "all"]:
|
|
self.assertFalse(weight.stop_gradient)
|
|
else:
|
|
self.assertTrue(weight.stop_gradient)
|
|
else:
|
|
if "bias" in name and bias == "all":
|
|
self.assertFalse(weight.stop_gradient)
|
|
else:
|
|
self.assertTrue(weight.stop_gradient)
|
|
input_ids = paddle.to_tensor(np.random.randint(100, 200, [1, 20]))
|
|
lorapro_model.train()
|
|
train_forward_results = lorapro_model(input_ids)
|
|
self.assertIsNotNone(train_forward_results)
|
|
lorapro_model.eval()
|
|
eval_forward_results = lorapro_model(input_ids)
|
|
self.assertIsNotNone(eval_forward_results)
|
|
self.assertTrue(paddle.allclose(train_forward_results[0], eval_forward_results[0]))
|
|
|
|
def test_lorapro_model_save_load(self):
|
|
with TemporaryDirectory() as tempdir:
|
|
input_ids = paddle.to_tensor(np.random.randint(100, 200, [1, 20]))
|
|
lorapro_config = LoRAConfig(target_modules=[".*q_proj.*", ".*v_proj.*"], r=4, lora_alpha=8, lorapro=True)
|
|
model = AutoModel.from_pretrained("__internal_testing__/tiny-random-bert")
|
|
lorapro_model = LoRAModel(model, lorapro_config)
|
|
lorapro_model.eval()
|
|
original_results = lorapro_model(input_ids)
|
|
lorapro_model.save_pretrained(tempdir)
|
|
|
|
loaded_lorapro_model = LoRAModel.from_pretrained(model, tempdir)
|
|
loaded_lorapro_model.eval()
|
|
loaded_results = loaded_lorapro_model(input_ids)
|
|
self.assertTrue(paddle.allclose(original_results[0], loaded_results[0]))
|
|
|
|
config_loaded_lorapro_model = LoRAModel.from_pretrained(model, tempdir, lora_config=lorapro_config)
|
|
config_loaded_lorapro_model.eval()
|
|
config_loaded_results = config_loaded_lorapro_model(input_ids)
|
|
self.assertTrue(paddle.allclose(original_results[0], config_loaded_results[0]))
|
|
|
|
@parameterized.expand([("zero",), ("sylvester",), ("symmetry",)])
|
|
def test_lorapro_modes(self, x_mode):
|
|
"""Test if AdamWLoRAPro optimizer with different x_modes can perform optimization steps"""
|
|
lorapro_config = LoRAConfig(
|
|
target_modules=[".*q_proj.*", ".*v_proj.*"],
|
|
r=4,
|
|
lora_alpha=8,
|
|
enable_lora_list=[None, [True, False]],
|
|
head_dim=2,
|
|
lorapro=True,
|
|
)
|
|
|
|
model = AutoModel.from_pretrained("__internal_testing__/tiny-random-bert")
|
|
lorapro_model = LoRAModel(model, lorapro_config)
|
|
lorapro_model.mark_only_lora_as_trainable()
|
|
|
|
input_ids = paddle.to_tensor(np.random.randint(100, 200, [2, 20]))
|
|
|
|
lorapro_model.train()
|
|
|
|
scaling_factor = lorapro_config.lora_alpha / lorapro_config.r
|
|
if lorapro_config.rslora:
|
|
scaling_factor = lorapro_config.lora_alpha / math.sqrt(lorapro_config.r)
|
|
|
|
optimizer = AdamWLoRAPro(
|
|
learning_rate=1e-4, parameters=lorapro_model.parameters(), scaling_factor=scaling_factor, x_mode=x_mode
|
|
)
|
|
|
|
outputs = lorapro_model(input_ids)
|
|
loss = outputs[0].mean()
|
|
|
|
loss.backward()
|
|
|
|
optimizer.step()
|
|
|
|
train_forward_results = lorapro_model(input_ids)
|
|
self.assertIsNotNone(train_forward_results)
|
|
self.assertIsInstance(optimizer, AdamWLoRAPro)
|
|
self.assertEqual(optimizer.x_mode, x_mode)
|
|
|
|
def test_lorapro_module_raise_exception(self):
|
|
lorapro_config = LoRAConfig(
|
|
target_modules=[".*norm1.*"], r=4, lora_alpha=8, enable_lora_list=None, lorapro=True
|
|
)
|
|
model = AutoModel.from_pretrained("__internal_testing__/tiny-random-bert")
|
|
with self.assertRaises(ValueError):
|
|
LoRAModel(model, lorapro_config)
|
|
|
|
|
|
class TestLoRAProConfig(unittest.TestCase):
|
|
def test_save_load(self):
|
|
with TemporaryDirectory() as tempdir:
|
|
lorapro_config = LoRAConfig()
|
|
lorapro_config.save_pretrained(tempdir)
|
|
loaded_lorapro_config = LoRAConfig.from_pretrained(tempdir)
|
|
self.assertEqual(lorapro_config, loaded_lorapro_config)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|