Spaces:
Sleeping
Sleeping
File size: 6,151 Bytes
6376749 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 |
# coding=utf-8
# Copyright 2023-present the HuggingFace Inc. team.
#
# 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 os
import tempfile
import unittest
import torch
from parameterized import parameterized
from transformers import AutoModelForCausalLM
from peft import (
PeftModel,
get_peft_model,
get_peft_model_state_dict,
prepare_model_for_int8_training,
)
from .testing_common import PeftTestConfigManager
# This has to be in the order: model_id, lora_kwargs, prefix_tuning_kwargs, prompt_encoder_kwargs, prompt_tuning_kwargs
PEFT_MODELS_TO_TEST = [
("hf-internal-testing/tiny-random-OPTForCausalLM", {"target_modules": ["q_proj", "v_proj"]}, {}, {}, {}),
]
class PeftTestMixin:
torch_device = "cuda" if torch.cuda.is_available() else "cpu"
class PeftModelTester(unittest.TestCase, PeftTestMixin):
r"""
Test if the PeftModel behaves as expected. This includes:
- test if the model has the expected methods
We use parametrized.expand for debugging purposes to test each model individually.
"""
@parameterized.expand(PeftTestConfigManager.get_grid_parameters(PEFT_MODELS_TO_TEST))
def test_attributes_parametrized(self, test_name, model_id, config_cls, config_kwargs):
self._test_model_attr(model_id, config_cls, config_kwargs)
def _test_model_attr(self, model_id, config_cls, config_kwargs):
model = AutoModelForCausalLM.from_pretrained(model_id)
config = config_cls(
base_model_name_or_path=model_id,
**config_kwargs,
)
model = get_peft_model(model, config)
self.assertTrue(hasattr(model, "save_pretrained"))
self.assertTrue(hasattr(model, "from_pretrained"))
self.assertTrue(hasattr(model, "push_to_hub"))
def _test_prepare_for_training(self, model_id, config_cls, config_kwargs):
model = AutoModelForCausalLM.from_pretrained(model_id).to(self.torch_device)
config = config_cls(
base_model_name_or_path=model_id,
**config_kwargs,
)
model = get_peft_model(model, config)
dummy_input = torch.LongTensor([[1, 1, 1]]).to(self.torch_device)
dummy_output = model.get_input_embeddings()(dummy_input)
self.assertTrue(not dummy_output.requires_grad)
# load with `prepare_model_for_int8_training`
model = AutoModelForCausalLM.from_pretrained(model_id).to(self.torch_device)
model = prepare_model_for_int8_training(model)
for param in model.parameters():
self.assertTrue(not param.requires_grad)
config = config_cls(
base_model_name_or_path=model_id,
**config_kwargs,
)
model = get_peft_model(model, config)
# For backward compatibility
if hasattr(model, "enable_input_require_grads"):
model.enable_input_require_grads()
else:
def make_inputs_require_grad(module, input, output):
output.requires_grad_(True)
model.get_input_embeddings().register_forward_hook(make_inputs_require_grad)
dummy_input = torch.LongTensor([[1, 1, 1]]).to(self.torch_device)
dummy_output = model.get_input_embeddings()(dummy_input)
self.assertTrue(dummy_output.requires_grad)
@parameterized.expand(PeftTestConfigManager.get_grid_parameters(PEFT_MODELS_TO_TEST))
def test_prepare_for_training_parametrized(self, test_name, model_id, config_cls, config_kwargs):
self._test_prepare_for_training(model_id, config_cls, config_kwargs)
def _test_save_pretrained(self, model_id, config_cls, config_kwargs):
model = AutoModelForCausalLM.from_pretrained(model_id)
config = config_cls(
base_model_name_or_path=model_id,
**config_kwargs,
)
model = get_peft_model(model, config)
model = model.to(self.torch_device)
with tempfile.TemporaryDirectory() as tmp_dirname:
model.save_pretrained(tmp_dirname)
model_from_pretrained = AutoModelForCausalLM.from_pretrained(model_id)
model_from_pretrained = PeftModel.from_pretrained(model_from_pretrained, tmp_dirname)
# check if the state dicts are equal
state_dict = get_peft_model_state_dict(model)
state_dict_from_pretrained = get_peft_model_state_dict(model_from_pretrained)
# check if same keys
self.assertEqual(state_dict.keys(), state_dict_from_pretrained.keys())
# check if tensors equal
for key in state_dict.keys():
self.assertTrue(
torch.allclose(
state_dict[key].to(self.torch_device), state_dict_from_pretrained[key].to(self.torch_device)
)
)
# check if `adapter_model.bin` is present
self.assertTrue(os.path.exists(os.path.join(tmp_dirname, "adapter_model.bin")))
# check if `adapter_config.json` is present
self.assertTrue(os.path.exists(os.path.join(tmp_dirname, "adapter_config.json")))
# check if `pytorch_model.bin` is not present
self.assertFalse(os.path.exists(os.path.join(tmp_dirname, "pytorch_model.bin")))
# check if `config.json` is not present
self.assertFalse(os.path.exists(os.path.join(tmp_dirname, "config.json")))
@parameterized.expand(PeftTestConfigManager.get_grid_parameters(PEFT_MODELS_TO_TEST))
def test_save_pretrained(self, test_name, model_id, config_cls, config_kwargs):
self._test_save_pretrained(model_id, config_cls, config_kwargs)
|