Hugging Face's logo

Spaces:
qinfeng722
/
llm-studio
Sleeping

App Files Community
llm-studio
File size: 8,636 Bytes 
5caedb4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
 
import os

import pytest
from jinja2.exceptions import TemplateError

from llm_studio.app_utils.hugging_face_utils import get_chat_template
from llm_studio.src.datasets.text_utils import get_tokenizer
from llm_studio.src.utils.config_utils import load_config_yaml


def test_chat_template_no_system_prompt():

    test_directory = os.path.abspath(os.path.dirname(__file__))
    cfg_path = os.path.join(test_directory, "../test_data/cfg.yaml")
    cfg = load_config_yaml(cfg_path)
    cfg.dataset.system_column = "None"
    cfg.dataset.text_prompt_start = "<|prompt|>"
    cfg.dataset.text_answer_separator = "<|answer|>"
    cfg.dataset.add_eos_token_to_prompt = True
    cfg.dataset.add_eos_token_to_answer = True

    tokenizer = get_tokenizer(cfg)
    tokenizer.eos_token = "</s>"
    tokenizer.chat_template = get_chat_template(cfg)

    chat = [
        {"role": "user", "content": "[user prompt]"},
        {"role": "assistant", "content": "[assistant response]"},
        {"role": "user", "content": "[user prompt2]"},
    ]

    input = tokenizer.apply_chat_template(
        chat,
        tokenize=False,
        add_generation_prompt=True,
    )
    expected = "<|prompt|>[user prompt]</s><|answer|>[assistant response]</s><|prompt|>[user prompt2]</s><|answer|>"  # noqa
    assert input == expected

    # raise error test
    for chat in [
        [
            {"role": "system", "content": "[system prompt]"},
            {"role": "user", "content": "[user prompt]"},
            {"role": "assistant", "content": "[assistant response]"},
            {"role": "user", "content": "[user prompt2]"},
        ],
        [
            {"role": "user", "content": "[user prompt]"},
            {"role": "assistant", "content": "[assistant response]"},
            {"role": "user", "content": "[user prompt2]"},
            {"role": "system", "content": "[system prompt]"},
        ],
    ]:
        with pytest.raises(TemplateError) as e:
            input = tokenizer.apply_chat_template(
                chat,
                tokenize=False,
                add_generation_prompt=True,
            )
        expected = "System role not supported"
        assert expected in str(e.value)

    # raise error test
    for chat in [
        [
            {"role": "user", "content": "[user prompt]"},
            {"role": "user", "content": "[user prompt2]"},
            {"role": "assistant", "content": "[assistant response]"},
        ],
        [
            {"role": "assistant", "content": "[assistant response]"},
            {"role": "user", "content": "[user prompt]"},
            {"role": "assistant", "content": "[assistant response]"},
        ],
        [
            {"role": "assistant", "content": "[assistant response]"},
            {"role": "assistant", "content": "[user prompt]"},
            {"role": "user", "content": "[assistant response]"},
        ],
    ]:
        with pytest.raises(TemplateError) as e:
            input = tokenizer.apply_chat_template(
                chat,
                tokenize=False,
                add_generation_prompt=True,
            )
        expected = "Conversation roles must alternate user/assistant/user/assistant/..."
        assert expected in str(e.value)


def test_chat_template_with_system_prompt():

    test_directory = os.path.abspath(os.path.dirname(__file__))
    cfg_path = os.path.join(test_directory, "../test_data/cfg.yaml")
    cfg = load_config_yaml(cfg_path)
    cfg.dataset.text_prompt_start = "<|prompt|>"
    cfg.dataset.text_answer_separator = "<|answer|>"
    cfg.dataset.text_system_start = "<|system|>"
    cfg.dataset.add_eos_token_to_prompt = True
    cfg.dataset.add_eos_token_to_answer = True
    cfg.dataset.add_eos_token_to_system = True
    cfg.dataset.system_column = "system"

    tokenizer = get_tokenizer(cfg)
    tokenizer.eos_token = "</s>"
    tokenizer.chat_template = get_chat_template(cfg)

    chat = [
        {"role": "system", "content": "[system prompt]"},
        {"role": "user", "content": "[user prompt]"},
        {"role": "assistant", "content": "[assistant response]"},
        {"role": "user", "content": "[user prompt2]"},
    ]

    input = tokenizer.apply_chat_template(
        chat,
        tokenize=False,
        add_generation_prompt=True,
    )
    expected = "<|system|>[system prompt]</s><|prompt|>[user prompt]</s><|answer|>[assistant response]</s><|prompt|>[user prompt2]</s><|answer|>"  # noqa
    assert input == expected

    # works w/o system prompt as well
    chat = [
        {"role": "user", "content": "[user prompt]"},
        {"role": "assistant", "content": "[assistant response]"},
        {"role": "user", "content": "[user prompt2]"},
    ]

    input = tokenizer.apply_chat_template(
        chat,
        tokenize=False,
        add_generation_prompt=True,
    )
    expected = "<|prompt|>[user prompt]</s><|answer|>[assistant response]</s><|prompt|>[user prompt2]</s><|answer|>"  # noqa
    assert input == expected

    # raise error test
    for chat in [
        [
            {"role": "user", "content": "[user prompt]"},
            {"role": "system", "content": "[system prompt]"},
            {"role": "user", "content": "[user prompt2]"},
            {"role": "assistant", "content": "[assistant response]"},
        ],
        [
            {"role": "system", "content": "[system prompt]"},
            {"role": "user", "content": "[user prompt]"},
            {"role": "user", "content": "[user prompt2]"},
            {"role": "assistant", "content": "[assistant response]"},
        ],
        [
            {"role": "user", "content": "[user prompt]"},
            {"role": "user", "content": "[user prompt2]"},
            {"role": "assistant", "content": "[assistant response]"},
        ],
        [
            {"role": "user", "content": "[user prompt]"},
            {"role": "assistant", "content": "[assistant response]"},
            {"role": "assistant", "content": "[assistant response2]"},
        ],
    ]:
        with pytest.raises(TemplateError) as e:
            input = tokenizer.apply_chat_template(
                chat,
                tokenize=False,
                add_generation_prompt=True,
            )
        expected = "Conversation roles must alternate system(optional)/user/assistant/user/assistant/..."  # noqa
        assert expected in str(e.value)


def test_chat_template_no_eos_token():

    test_directory = os.path.abspath(os.path.dirname(__file__))
    cfg_path = os.path.join(test_directory, "../test_data/cfg.yaml")
    cfg = load_config_yaml(cfg_path)
    cfg.dataset.text_prompt_start = "<|prompt|>"
    cfg.dataset.text_answer_separator = "<|answer|>"
    cfg.dataset.text_system_start = "<|system|>"
    cfg.dataset.add_eos_token_to_system = False
    cfg.dataset.add_eos_token_to_prompt = False
    cfg.dataset.add_eos_token_to_answer = False
    cfg.dataset.system_column = "system"

    tokenizer = get_tokenizer(cfg)
    tokenizer.chat_template = get_chat_template(cfg)

    chat = [
        {"role": "system", "content": "[system prompt]"},
        {"role": "user", "content": "[user prompt]"},
        {"role": "assistant", "content": "[assistant response]"},
        {"role": "user", "content": "[user prompt2]"},
    ]

    input = tokenizer.apply_chat_template(
        chat,
        tokenize=False,
        add_generation_prompt=True,
    )
    expected = "<|system|>[system prompt]<|prompt|>[user prompt]<|answer|>[assistant response]<|prompt|>[user prompt2]<|answer|>"  # noqa
    assert input == expected


def test_chat_template_no_special_token():

    test_directory = os.path.abspath(os.path.dirname(__file__))
    cfg_path = os.path.join(test_directory, "../test_data/cfg.yaml")
    cfg = load_config_yaml(cfg_path)
    cfg.dataset.system_column = "system"
    cfg.dataset.text_system_start = ""
    cfg.dataset.text_prompt_start = ""
    cfg.dataset.text_answer_separator = ""
    cfg.dataset.add_eos_token_to_system = False
    cfg.dataset.add_eos_token_to_prompt = False
    cfg.dataset.add_eos_token_to_answer = False

    tokenizer = get_tokenizer(cfg)
    tokenizer.chat_template = get_chat_template(cfg)

    chat = [
        {"role": "system", "content": "[system prompt]"},
        {"role": "user", "content": "[user prompt]"},
        {"role": "assistant", "content": "[assistant response]"},
        {"role": "user", "content": "[user prompt2]"},
    ]

    input = tokenizer.apply_chat_template(
        chat,
        tokenize=False,
        add_generation_prompt=True,
    )
    expected = "[system prompt][user prompt][assistant response][user prompt2]"
    assert input == expected