File size: 16,276 Bytes
cfcc549
e6fdeb0
 
 
7bbaac9
e65aeed
a363604
e6fdeb0
a363604
3392270
e6fdeb0
7925ddc
3a38271
 
 
 
3392270
 
 
 
7925ddc
 
 
 
e7d3e2d
553a86b
 
e6fdeb0
7bbaac9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e6fdeb0
 
cfcc549
 
 
 
a363604
 
 
 
 
 
e6fdeb0
0136f51
e6fdeb0
 
 
 
 
 
 
 
 
 
cfcc549
 
 
e6fdeb0
 
cfcc549
 
 
 
e6fdeb0
 
e7d3e2d
e6fdeb0
 
 
 
 
 
 
 
 
 
 
a363604
 
 
 
 
 
 
e7d3e2d
a363604
 
 
 
 
 
 
 
 
 
f30afe4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5ada140
 
7bbaac9
5ada140
 
7bbaac9
5ada140
 
 
 
 
7bbaac9
 
5ada140
 
 
7bbaac9
 
5ada140
 
 
 
7bbaac9
5ada140
 
 
 
7bbaac9
5ada140
 
 
 
7bbaac9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f30afe4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1ab3bf3
7925ddc
 
 
 
baed440
7925ddc
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
baed440
7925ddc
 
 
 
 
 
 
 
 
 
 
 
 
e6fdeb0
3a38271
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2bb0b78
 
 
 
 
 
 
 
 
3a38271
 
3392270
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e6fdeb0
 
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
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
"""Module for testing prompt tokenizers."""
import json
import logging
import unittest
from copy import deepcopy
from pathlib import Path
from typing import Optional

import pytest
from transformers import AutoTokenizer, LlamaTokenizer

from axolotl.prompt_strategies.alpaca_chat import NoSystemPrompter
from axolotl.prompt_strategies.alpaca_w_system import (
    InstructionWSystemPromptTokenizingStrategy,
    SystemDataPrompter,
)
from axolotl.prompt_strategies.llama2_chat import (
    Llama2ChatPrompter,
    LLama2ChatTokenizingStrategy,
)
from axolotl.prompt_tokenizers import (
    AlpacaPromptTokenizingStrategy,
    ShareGPTPromptTokenizingStrategy,
)
from axolotl.prompters import AlpacaPrompter, PromptStyle, ShareGPTPrompterV2

LOG = logging.getLogger("axolotl")

test_data = {
    "multi_turn_sys": {
        "conversations": [
            {"from": "system", "value": "lorem"},
            {"from": "human", "value": "abc"},
            {"from": "gpt", "value": "ipsum"},
            {"from": "human", "value": "123"},
            {"from": "gpt", "value": "sit"},
        ]
    },
    "single_turn_sys": {
        "conversations": [
            {"from": "system", "value": "lorem"},
            {"from": "human", "value": "abc"},
            {"from": "gpt", "value": "ipsum"},
        ]
    },
    "single_turn_no_sys": {
        "conversations": [
            {"from": "human", "value": "abc"},
            {"from": "gpt", "value": "ipsum"},
        ]
    },
    "multi_turn_no_sys": {
        "conversations": [
            {"from": "human", "value": "abc"},
            {"from": "gpt", "value": "ipsum"},
            {"from": "human", "value": "123"},
            {"from": "gpt", "value": "sit"},
        ]
    },
}


def prompt_strat(conversation, tokenizer):
    "Helper function to create a prompt strategy for testing."
    prompter = ShareGPTPrompterV2(conversation=conversation)
    return ShareGPTPromptTokenizingStrategy(
        prompter,
        tokenizer,
        False,
        2048,
    )


class TestPromptTokenizationStrategies(unittest.TestCase):
    """
    Test class for prompt tokenization strategies.
    """

    _caplog: Optional[pytest.LogCaptureFixture] = None

    @pytest.fixture(autouse=True)
    def inject_fixtures(self, caplog):
        self._caplog = caplog

    def setUp(self) -> None:
        # pylint: disable=duplicate-code
        self.tokenizer = AutoTokenizer.from_pretrained("huggyllama/llama-7b")
        self.tokenizer.add_special_tokens(
            {
                "bos_token": "<s>",
                "eos_token": "</s>",
                "unk_token": "<unk>",
            }
        )

    def test_sharegpt_integration(self):
        with open(
            Path(__file__).parent / "fixtures/conversation.json", encoding="utf-8"
        ) as fin:
            data = fin.read()
            conversation = json.loads(data)
        with open(
            Path(__file__).parent / "fixtures/conversation.tokenized.json",
            encoding="utf-8",
        ) as fin:
            data = fin.read()
            tokenized_conversation = json.loads(data)
        prompter = ShareGPTPrompterV2()
        strat = ShareGPTPromptTokenizingStrategy(
            prompter,
            self.tokenizer,
            False,
            2048,
        )
        example = strat.tokenize_prompt(conversation)
        for fields in ["input_ids", "attention_mask", "labels"]:
            self.assertEqual(len(example[fields]), len(tokenized_conversation[fields]))
            self.assertEqual(example[fields], tokenized_conversation[fields])

    def test_sharegpt_warnings_integration(self):
        with open(
            Path(__file__).parent / "fixtures/conversation.missingturns.json",
            encoding="utf-8",
        ) as fin:
            data = fin.read()
            conversation = json.loads(data)
        prompter = ShareGPTPrompterV2()
        strat = ShareGPTPromptTokenizingStrategy(
            prompter,
            self.tokenizer,
            False,
            2048,
        )
        with self._caplog.at_level(logging.WARNING):
            strat.tokenize_prompt(conversation)
            assert "assistant turn has empty text" in self._caplog.records[1].message

    def test_sharegpt_warnings_turns(self):
        conversation = {
            "conversations": [
                {"from": "system", "value": "lorem"},
                {"from": "gpt", "value": "ipsum"},
                {"from": "human", "value": "dolor"},
                {"from": "human", "value": "dolor"},
                {"from": "gpt", "value": "sit"},
            ]
        }
        prompter = ShareGPTPrompterV2()
        strat = ShareGPTPromptTokenizingStrategy(
            prompter,
            self.tokenizer,
            False,
            2048,
        )
        with self._caplog.at_level(logging.WARNING):
            strat.tokenize_prompt(conversation)
            assert (
                "Role did not alternate between turns (gpt and human)"
                in self._caplog.records[0].message
            )

    def test_sharegpt_llama(self):
        "Make sure the sharegpt/llama is tokenized and formatted correctly."
        strat = prompt_strat("llama-2", self.tokenizer)

        def tokenize(conv):
            return strat.tokenize_prompt(deepcopy(conv))["input_ids"]

        def decode(ids):
            return strat.tokenizer.decode(ids)

        # fmt: off
        # System message, multi-turn conversations
        mt_ids = tokenize(test_data['multi_turn_sys'])
        assert decode(mt_ids) == '<s> [INST] <<SYS>>\nlorem\n<</SYS>>\n\nabc [/INST] ipsum</s><s> [INST] 123 [/INST] sit</s>'
        assert mt_ids == [1, 518, 25580, 29962, 3532, 14816, 29903, 6778, 13, 29880, 3668, 13, 29966, 829, 14816, 29903, 6778, 13, 13, 10736, 518, 29914, 25580, 29962, 23421, 2, 1, 518, 25580, 29962, 29871, 29896, 29906, 29941, 518, 29914, 25580, 29962, 7845, 2]

        # System message, single-turn conversations
        st_ids = tokenize(test_data['single_turn_sys'])
        assert decode(st_ids) == '<s> [INST] <<SYS>>\nlorem\n<</SYS>>\n\nabc [/INST] ipsum</s>'
        assert st_ids == [1, 518, 25580, 29962, 3532, 14816, 29903, 6778, 13, 29880, 3668, 13, 29966, 829, 14816, 29903, 6778, 13, 13, 10736, 518, 29914, 25580, 29962, 23421, 2]

        # No system message, single-turn
        ns_ids = tokenize(test_data['single_turn_no_sys'])
        assert decode(ns_ids) == '<s> [INST] abc [/INST] ipsum</s>'
        assert ns_ids == [1, 518, 25580, 29962, 25638, 518, 29914, 25580, 29962, 23421, 2]

        # No system message, multi-turn
        ns_mt_ids = tokenize(test_data['multi_turn_no_sys'])
        assert decode(ns_mt_ids) == '<s> [INST] abc [/INST] ipsum</s><s> [INST] 123 [/INST] sit</s>'
        assert ns_mt_ids == [1, 518, 25580, 29962, 25638, 518, 29914, 25580, 29962, 23421, 2, 1, 518, 25580, 29962, 29871, 29896, 29906, 29941, 518, 29914, 25580, 29962, 7845, 2]
        # fmt: on

    def test_sharegpt_mistral(self):
        "Make sure the sharegpt/mistral is tokenized and formatted correctly."
        strat = prompt_strat("mistral", self.tokenizer)

        def tokenize(conv):
            return strat.tokenize_prompt(deepcopy(conv))["input_ids"]

        def decode(ids):
            return strat.tokenizer.decode(ids)

        # fmt: off
        # System message, multi-turn conversations
        mt_ids = tokenize(test_data['multi_turn_sys'])
        assert decode(mt_ids) == '<s> [INST] lorem\nabc [/INST] ipsum</s> [INST] 123 [/INST] sit</s>'
        assert mt_ids == [1, 518, 25580, 29962, 301, 3668, 13, 10736, 518, 29914, 25580, 29962, 23421, 2, 518, 25580, 29962, 29871, 29896, 29906, 29941, 518, 29914, 25580, 29962, 7845, 2]

        # System message, single-turn conversations
        st_ids = tokenize(test_data['single_turn_sys'])
        assert decode(st_ids) == '<s> [INST] lorem\nabc [/INST] ipsum</s>'
        assert st_ids == [1, 518, 25580, 29962, 301, 3668, 13, 10736, 518, 29914, 25580, 29962, 23421, 2]

        # No system message, single-turn
        ns_ids = tokenize(test_data['single_turn_no_sys'])
        assert decode(ns_ids) == '<s> [INST] abc [/INST] ipsum</s>'
        assert ns_ids == [1, 518, 25580, 29962, 25638, 518, 29914, 25580, 29962, 23421, 2]

        # No system message, multi-turn
        ns_mt_ids = tokenize(test_data['multi_turn_no_sys'])
        assert decode(ns_mt_ids) == '<s> [INST] abc [/INST] ipsum</s> [INST] 123 [/INST] sit</s>'
        assert ns_mt_ids == [1, 518, 25580, 29962, 25638, 518, 29914, 25580, 29962, 23421, 2, 518, 25580, 29962, 29871, 29896, 29906, 29941, 518, 29914, 25580, 29962, 7845, 2]
        # fmt: on

    def test_sharegpt_changes_roles(self):
        conversation = {
            "roles": ["USER", "CHARACTER"],
            "conversations": [
                {"from": "system", "value": "lorem"},
                {"from": "gpt", "value": "ipsum"},
                {"from": "human", "value": "dolor"},
                {"from": "gpt", "value": "sit"},
            ],
        }
        prompter = ShareGPTPrompterV2()
        strat = ShareGPTPromptTokenizingStrategy(
            prompter,
            self.tokenizer,
            False,
            2048,
        )
        with self._caplog.at_level(logging.WARNING):
            res = strat.tokenize_prompt(conversation)
            assert "CHARACTER" in self.tokenizer.decode(res["input_ids"])

    def test_sharegpt_assistant_label_ignore(self):
        conversation = {
            "roles": ["user", "assistant"],
            "conversations": [
                {"from": "system", "value": "lorem"},
                {"from": "gpt", "value": "ipsum"},
                {"from": "human", "value": "dolor"},
                {"from": "gpt", "value": "sit"},
            ],
        }
        prompter = ShareGPTPrompterV2()
        strat = ShareGPTPromptTokenizingStrategy(
            prompter,
            self.tokenizer,
            False,
            2048,
        )
        with self._caplog.at_level(logging.WARNING):
            res = strat.tokenize_prompt(conversation)
            idx = res["input_ids"].index(20255)  # assistant token
            assert res["labels"][idx] == -100

    def test_no_sys_prompt(self):
        """
        tests the interface between the user and assistant parts
        """
        prompter = NoSystemPrompter()
        # pylint: disable=duplicate-code
        strat = AlpacaPromptTokenizingStrategy(
            prompter,
            self.tokenizer,
            False,
            2048,
        )
        sample = {
            "instruction": "hello cruel. lorem ipsum dolor sit amet.",
            "output": "world!",
        }
        example = strat.tokenize_prompt(sample)
        world_idx = example["input_ids"].index(3186)
        assert example["labels"][world_idx] == 3186
        assert example["labels"][world_idx - 1] == -100

    def test_alpaca(self):
        """
        tests the interface between the user and assistant parts
        """
        # pylint: disable=duplicate-code
        prompter = AlpacaPrompter()
        strat = AlpacaPromptTokenizingStrategy(
            prompter,
            self.tokenizer,
            False,
            2048,
        )
        sample = {"instruction": "hello!", "output": "Hi! How can I help?"}
        example = strat.tokenize_prompt(sample)
        world_idx = example["input_ids"].index(6324)
        assert example["labels"][world_idx] == 6324
        assert example["labels"][world_idx - 1] == -100


class InstructionWSystemPromptTokenizingStrategyTest(unittest.TestCase):
    """
    Test class for prompt tokenization strategies with sys prompt from the dataset
    """

    def setUp(self) -> None:
        # pylint: disable=duplicate-code
        self.tokenizer = AutoTokenizer.from_pretrained("huggyllama/llama-7b")
        self.tokenizer.add_special_tokens(
            {
                "bos_token": "<s>",
                "eos_token": "</s>",
                "unk_token": "<unk>",
            }
        )

    def test_system_alpaca(self):
        prompter = SystemDataPrompter(PromptStyle.CHAT.value)
        strat = InstructionWSystemPromptTokenizingStrategy(
            prompter,
            self.tokenizer,
            False,
            2048,
        )
        sample = {
            "system": "use cot",
            "instruction": "hello!",
            "output": "Hi! How can I help?",
        }
        example = strat.tokenize_prompt(sample)
        assert example["input_ids"][0:5] == [
            1,
            28962,
            1254,
            12665,
            29901,
        ]  # "<s>SYSTEM:"
        assert example["input_ids"][5:7] == [671, 20118]  # " use cot"
        assert example["input_ids"][8] == 11889  # USER


class Llama2ChatTokenizationTest(unittest.TestCase):
    """
    Test class for prompt tokenization strategies with sys prompt from the dataset
    """

    def setUp(self) -> None:
        # pylint: disable=duplicate-code
        self.tokenizer = LlamaTokenizer.from_pretrained("NousResearch/Llama-2-7b-hf")
        # woraround because official Meta repos are not open

    def test_llama2_chat_integration(self):
        with open(
            Path(__file__).parent / "fixtures/conversation.json", encoding="utf-8"
        ) as fin:
            data = fin.read()
            conversation = json.loads(data)
        with open(
            Path(__file__).parent / "fixtures/conversation.tokenized_llama2chat.json",
            encoding="utf-8",
        ) as fin:
            data = fin.read()
            tokenized_conversation = json.loads(data)
        prompter = Llama2ChatPrompter()
        strat = LLama2ChatTokenizingStrategy(
            prompter,
            self.tokenizer,
            False,
            4096,
        )
        example = strat.tokenize_prompt(conversation)
        for fields in ["input_ids", "attention_mask", "labels"]:
            self.assertEqual(len(example[fields]), len(tokenized_conversation[fields]))
            self.assertEqual(example[fields], tokenized_conversation[fields])

    def compare_with_transformers_integration(self):
        # this needs transformers >= v4.31.0
        from transformers.models.llama.tokenization_llama import B_SYS, E_SYS
        from transformers.pipelines.conversational import Conversation

        # from transformers.models.llama.tokenization_llama import DEFAULT_SYSTEM_PROMPT
        # broken as of 23/7/20
        # see https://github.com/huggingface/transformers/pull/24935
        # pylint: disable=C0103
        DEFAULT_SYSTEM_PROMPT = """\
You are a helpful, respectful and honest assistant. Always answer as helpfully as possible, while being safe. Your answers should not include any harmful, unethical, racist, sexist, toxic, dangerous, or illegal content. Please ensure that your responses are socially unbiased and positive in nature.

If a question does not make any sense, or is not factually coherent, explain why instead of answering something not correct. If you don't know the answer to a question, please don't share false information."""
        with open(
            Path(__file__).parent / "fixtures/conversation.json", encoding="utf-8"
        ) as fin:
            data = fin.read()
            conversation = json.loads(data)
        with open(
            Path(__file__).parent / "fixtures/conversation.tokenized_llama2chat.json",
            encoding="utf-8",
        ) as fin:
            data = fin.read()
            tokenized_conversation = json.loads(data)

        user_input = []
        answers = []
        for msg in conversation["conversations"]:
            if msg["from"] == "human":
                user_input.append(msg["value"])
            else:
                answers.append(msg["value"])
        hf_conf = Conversation(
            text=user_input[-1],
            past_user_inputs=[B_SYS + DEFAULT_SYSTEM_PROMPT + E_SYS + user_input[0]]
            + user_input[1:-1],
            generated_responses=answers,
        )
        # pylint: disable=W0212
        hf_tokens = self.tokenizer._build_conversation_input_ids(hf_conf)

        self.assertEqual(
            hf_tokens, tokenized_conversation["input_ids"][: len(hf_tokens)]
        )


if __name__ == "__main__":
    unittest.main()