File size: 8,855 Bytes
b381930
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
# Copyright 2025 The HuggingFace Team. 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 gc
import tempfile
import unittest

from transformers import AutoModelForCausalLM, AutoRoundConfig, AutoTokenizer
from transformers.testing_utils import (
    require_accelerate,
    require_auto_round,
    require_intel_extension_for_pytorch,
    require_torch_gpu,
    require_torch_multi_gpu,
    slow,
    torch_device,
)
from transformers.utils import is_torch_available


if is_torch_available():
    import torch


@slow
@require_torch_gpu
@require_auto_round
@require_accelerate
class AutoRoundTest(unittest.TestCase):
    model_name = "OPEA/Qwen2.5-1.5B-Instruct-int4-sym-inc"
    input_text = "There is a girl who likes adventure,"
    EXPECTED_OUTPUTS = set()
    ## Different backends may produce slight variations in output
    EXPECTED_OUTPUTS.add(
        "There is a girl who likes adventure, and she has been exploring the world "
        "for many years. She travels to different countries and cultures, trying new "
        "things every day. One of her favorite places to visit is a small village in "
        "the mountains where"
    )
    EXPECTED_OUTPUTS.add(
        "There is a girl who likes adventure, and she has been exploring the world for many years. She has visited every country in Europe and has even traveled to some of the most remote parts of Africa. She enjoys hiking through the mountains and discovering"
    )

    device_map = "cuda"

    # called only once for all test in this class
    @classmethod
    def setUpClass(cls):
        """
        Setup quantized model
        """
        torch.cuda.synchronize()
        cls.tokenizer = AutoTokenizer.from_pretrained(cls.model_name)
        cls.quantized_model = AutoModelForCausalLM.from_pretrained(
            cls.model_name, device_map=cls.device_map, torch_dtype=torch.float16
        )

    def tearDown(self):
        gc.collect()
        torch.cuda.empty_cache()
        gc.collect()

    def test_quantized_model(self):
        """
        Simple test that checks if the quantized model is working properly
        """
        input_ids = self.tokenizer(self.input_text, return_tensors="pt").to(torch_device)
        output = self.quantized_model.generate(**input_ids, max_new_tokens=40, do_sample=False)
        self.assertIn(self.tokenizer.decode(output[0], skip_special_tokens=True), self.EXPECTED_OUTPUTS)

    def test_raise_if_non_quantized(self):
        model_id = "facebook/opt-125m"
        quantization_config = AutoRoundConfig(bits=4)
        with self.assertRaises(ValueError):
            _ = AutoModelForCausalLM.from_pretrained(model_id, quantization_config=quantization_config)

    def test_quantized_model_bf16(self):
        """
        Simple test that checks if the quantized model is working properly with bf16
        """
        input_ids = self.tokenizer(self.input_text, return_tensors="pt").to(torch_device)
        quantization_config = AutoRoundConfig(backend="triton")
        quantized_model = AutoModelForCausalLM.from_pretrained(
            self.model_name,
            torch_dtype=torch.bfloat16,
            device_map=self.device_map,
            quantization_config=quantization_config,
        )

        output = quantized_model.generate(**input_ids, max_new_tokens=40, do_sample=False)
        self.assertIn(self.tokenizer.decode(output[0], skip_special_tokens=True), self.EXPECTED_OUTPUTS)

    @require_intel_extension_for_pytorch
    def test_quantized_model_on_cpu(self):
        """
        Simple test that checks if the quantized model is working properly
        """
        input_ids = self.tokenizer(self.input_text, return_tensors="pt")

        quantized_model = AutoModelForCausalLM.from_pretrained(self.model_name, torch_dtype="auto")
        output = quantized_model.generate(**input_ids, max_new_tokens=40, do_sample=False)

        self.assertIn(self.tokenizer.decode(output[0], skip_special_tokens=True), self.EXPECTED_OUTPUTS)

    def test_save_pretrained(self):
        """
        Simple test that checks if the quantized model is working properly after being saved and loaded
        """

        ## some backends like marlin/ipex will repack the weight that caused the weight shape changed
        with tempfile.TemporaryDirectory() as tmpdirname:
            quantization_config = AutoRoundConfig(backend="triton")
            quantized_model = AutoModelForCausalLM.from_pretrained(
                self.model_name,
                device_map=self.device_map,
                torch_dtype=torch.float16,
                quantization_config=quantization_config,
            )

            quantized_model.save_pretrained(tmpdirname)
            model = AutoModelForCausalLM.from_pretrained(tmpdirname, device_map="cuda")

            input_ids = self.tokenizer(self.input_text, return_tensors="pt").to(torch_device)

            output = model.generate(**input_ids, max_new_tokens=40, do_sample=False)
            self.assertIn(self.tokenizer.decode(output[0], skip_special_tokens=True), self.EXPECTED_OUTPUTS)

    @require_torch_multi_gpu
    def test_quantized_model_multi_gpu(self):
        """
        Simple test that checks if the quantized model is working properly with multiple GPUs
        """
        input_ids = self.tokenizer(self.input_text, return_tensors="pt").to(torch_device)
        quantization_config = AutoRoundConfig(backend="triton")
        quantized_model = AutoModelForCausalLM.from_pretrained(
            self.model_name, device_map="auto", quantization_config=quantization_config, torch_dtype="auto"
        )

        output = quantized_model.generate(**input_ids, max_new_tokens=40, do_sample=False)

        self.assertIn(self.tokenizer.decode(output[0], skip_special_tokens=True), self.EXPECTED_OUTPUTS)

    def test_convert_from_gptq(self):
        """
        Simple test that checks if auto-round work properly wth gptq format
        """
        model_name = "ybelkada/opt-125m-gptq-4bit"

        quantization_config = AutoRoundConfig()

        model = AutoModelForCausalLM.from_pretrained(
            model_name, device_map="cuda", quantization_config=quantization_config, torch_dtype="auto"
        )
        tokenizer = AutoTokenizer.from_pretrained(model_name)

        text = "There is a girl who likes adventure,"
        inputs = tokenizer(text, return_tensors="pt").to(model.device)
        tokenizer.decode(model.generate(**inputs, max_new_tokens=5)[0])

    @require_intel_extension_for_pytorch
    def test_convert_from_awq_cpu(self):
        """
        Simple test that checks if auto-round work properly wth awq format
        """
        model_name = "casperhansen/opt-125m-awq"

        quantization_config = AutoRoundConfig()

        model = AutoModelForCausalLM.from_pretrained(
            model_name, device_map="cpu", quantization_config=quantization_config, torch_dtype="auto"
        )
        tokenizer = AutoTokenizer.from_pretrained(model_name)

        text = "There is a girl who likes adventure,"
        inputs = tokenizer(text, return_tensors="pt").to(model.device)
        tokenizer.decode(model.generate(**inputs, max_new_tokens=5)[0])

    def test_mixed_bits(self):
        """
        Simple test that checks if auto-round work properly wth mixed bits
        """
        model_name = "facebook/opt-125m"
        model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype="auto")
        tokenizer = AutoTokenizer.from_pretrained(model_name)
        layer_config = {
            "model.decoder.layers.0.self_attn.k_proj": {"bits": 8},
            "model.decoder.layers.6.self_attn.out_proj": {"bits": 2, "group_size": 32},
        }

        bits, group_size, sym = 4, 128, True
        from auto_round import AutoRound

        autoround = AutoRound(model, tokenizer, bits=bits, group_size=group_size, sym=sym, layer_config=layer_config)
        with tempfile.TemporaryDirectory() as tmpdirname:
            autoround.quantize_and_save(output_dir=tmpdirname)
            model = AutoModelForCausalLM.from_pretrained(tmpdirname, torch_dtype=torch.float16, device_map="cuda")
            text = "There is a girl who likes adventure,"
            inputs = tokenizer(text, return_tensors="pt").to(model.device)
            tokenizer.decode(model.generate(**inputs, max_new_tokens=5)[0])