File size: 10,217 Bytes
6fcd376
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from dataclasses import dataclass, field
from typing import Dict, Optional, Sequence
import logging
import os, sys
import copy

import torch
import transformers
from transformers import LlamaForCausalLM, LlamaTokenizer

from torch.utils.data import Dataset
from transformers import Trainer

sys.path.append(os.path.dirname(__file__))
sys.path.append(os.path.dirname(os.path.dirname(__file__)))
from utils.special_tok_llama2 import (
    B_CODE,
    E_CODE,
    B_RESULT,
    E_RESULT,
    B_INST,
    E_INST,
    B_SYS,
    E_SYS,
    DEFAULT_PAD_TOKEN,
    DEFAULT_BOS_TOKEN,
    DEFAULT_EOS_TOKEN,
    DEFAULT_UNK_TOKEN,
    IGNORE_INDEX,
)

from conversation_template import json_to_code_result_tok_temp


@dataclass
class ModelArguments:
    model_name_or_path: Optional[str] = field(default="./ckpt/llama-2-13b-chat")
    peft: bool = field(default=False)


@dataclass
class DataArguments:
    data_path: str = field(
        default=None, metadata={"help": "Path to the training data."}
    )


@dataclass
class TrainingArguments(transformers.TrainingArguments):
    cache_dir: Optional[str] = field(default=None)
    optim: str = field(default="adamw_torch")
    model_max_length: int = field(
        default=4096,
        metadata={
            "help": "Maximum sequence length. Sequences will be right padded (and possibly truncated)."
        },
    )


def create_peft_config(model):
    from peft import (
        get_peft_model,
        LoraConfig,
        TaskType,
        prepare_model_for_int8_training,
    )

    peft_config = LoraConfig(
        task_type=TaskType.CAUSAL_LM,
        inference_mode=False,
        r=8,
        lora_alpha=16,
        lora_dropout=0.05,
        target_modules=["q_proj", "v_proj"],
    )

    # prepare int-8 model for training
    model = prepare_model_for_int8_training(model)
    model = get_peft_model(model, peft_config)
    model.print_trainable_parameters()
    print(f"Using Peft")
    return model, peft_config


def _tokenize_fn(
    strings: Sequence[str], tokenizer: transformers.PreTrainedTokenizer
) -> Dict:
    """Tokenize a list of strings."""
    tokenized_list = [
        tokenizer(
            text,
            return_tensors="pt",
            padding="longest",
            max_length=tokenizer.model_max_length,
            truncation=True,
        )
        for text in strings
    ]

    input_ids = [tokenized.input_ids[0] for tokenized in tokenized_list]
    input_ids_lens = [
        tokenized.input_ids.ne(tokenizer.pad_token_id).sum().item()
        for tokenized in tokenized_list
    ]
    return dict(
        input_ids=input_ids,
        input_ids_lens=input_ids_lens,
    )


def find_all_sublist_end(main_list, sublist):
    """Find all the ending indices of a sublist in a main list."""
    sublist_len = len(sublist)
    main_list = main_list.tolist()
    indices = []
    for index in (i for i, e in enumerate(main_list) if e == sublist[0]):
        if main_list[index : index + sublist_len] == sublist:
            indices.append(index + sublist_len)
    return indices


def find_all_sublist_start(main_list, sublist):
    """Find all the starting indices of a sublist in a main list."""
    sublist_len = len(sublist)
    main_list = main_list.tolist()
    indices = []
    for index in (i for i, e in enumerate(main_list) if e == sublist[0]):
        if main_list[index : index + sublist_len] == sublist:
            indices.append(index)
    return indices


def preprocess(
    trajs: Sequence[str],
    tokenizer: transformers.PreTrainedTokenizer,
) -> Dict:
    INST_START_INDEX = tokenizer.encode(f"{B_INST}")[-1]
    INST_END_INDEX = tokenizer.encode(f"{E_INST}")[-1]
    RESULT_START_INDEX = tokenizer.encode(f"{B_RESULT}")[-1]
    RESULT_END_INDEX = tokenizer.encode(f"{E_RESULT}")[-1]

    """Preprocess the data by tokenizing."""
    examples_tokenized = _tokenize_fn(trajs, tokenizer)

    input_ids_lens = examples_tokenized["input_ids_lens"]
    input_ids = examples_tokenized["input_ids"]  # [torch.tensor , torch.tensor , ...]
    labels = copy.deepcopy(input_ids)

    # IGNORE INDEX SET
    for i, label in enumerate(labels):
        user_start_inds = find_all_sublist_start(label, [INST_START_INDEX])
        assistant_start_inds = find_all_sublist_end(label, [INST_END_INDEX])

        result_start_inds = find_all_sublist_start(label, [RESULT_START_INDEX])
        result_end_inds = find_all_sublist_end(label, [RESULT_END_INDEX])

        # for debug
        # for len_i, ind in enumerate(label):
        #    print(f'{len_i}|{ind} -> "{tokenizer.decode(ind)}"')

        assert len(user_start_inds) == len(
            assistant_start_inds
        ), f"User and Assistant pair should be equal :: \n\tUser [{user_start_inds}]/\n\tAssistant [{assistant_start_inds}]\n\n Text : \n{trajs[i]}"

        assert len(result_start_inds) == len(
            result_end_inds
        ), f"Start and End indices pairs do not match.: : \nText : \n{trajs[i]}"

        for user_start_ind, assistant_start_ind in zip(
            user_start_inds, assistant_start_inds
        ):
            label[user_start_ind + 1 : assistant_start_ind - 1] = IGNORE_INDEX

        for start, end in zip(result_start_inds, result_end_inds):
            label[start + 1 : end - 1] = IGNORE_INDEX

    # cut max length
    input_ids = [i[:1500] for i in input_ids]
    labels = [i[:1500] for i in labels]

    return dict(input_ids=input_ids, labels=labels)


class SupervisedDataset(Dataset):
    """Dataset for supervised fine-tuning."""

    def __init__(self, data_path: str, tokenizer: transformers.PreTrainedTokenizer):
        super(SupervisedDataset, self).__init__()
        logging.warning(f"Loading data from data path : {data_path}")
        all_json = os.listdir(data_path)

        trajs = list()
        for json_file_name in all_json:
            traj = json_to_code_result_tok_temp(json_file_name=json_file_name)
            trajs.append(traj)

        logging.warning("Tokenizing inputs... This may take some time...")
        data_dict = preprocess(trajs, tokenizer)

        self.input_ids = data_dict["input_ids"]
        self.labels = data_dict["labels"]

    def __len__(self):
        return len(self.input_ids)

    def __getitem__(self, i) -> Dict[str, torch.Tensor]:
        return dict(input_ids=self.input_ids[i], labels=self.labels[i])


@dataclass
class DataCollatorForSupervisedDataset(object):
    """Collate examples for supervised fine-tuning."""

    tokenizer: transformers.PreTrainedTokenizer

    def __call__(self, instances: Sequence[Dict]) -> Dict[str, torch.Tensor]:
        input_ids, labels = tuple(
            [instance[key] for instance in instances] for key in ("input_ids", "labels")
        )
        input_ids = torch.nn.utils.rnn.pad_sequence(
            input_ids, batch_first=True, padding_value=self.tokenizer.pad_token_id
        )
        labels = torch.nn.utils.rnn.pad_sequence(
            labels, batch_first=True, padding_value=IGNORE_INDEX
        )
        return dict(
            input_ids=input_ids,
            labels=labels,
            attention_mask=input_ids.ne(self.tokenizer.pad_token_id),
        )


def make_supervised_data_module(
    tokenizer: transformers.PreTrainedTokenizer, data_args
) -> Dict:
    """Make dataset and collator for supervised fine-tuning."""
    train_dataset = SupervisedDataset(
        tokenizer=tokenizer, data_path=data_args.data_path
    )
    data_collator = DataCollatorForSupervisedDataset(tokenizer=tokenizer)
    return dict(
        train_dataset=train_dataset, eval_dataset=None, data_collator=data_collator
    )


def build_model_from_hf_path(
    hf_model_path: str = "./ckpt/llama-2-13b-chat", peft: bool = False
):
    # build tokenizer
    tokenizer = LlamaTokenizer.from_pretrained(
        hf_model_path,
        padding_side="right",
        use_fast=False,
    )

    special_tokens_dict = dict()
    if tokenizer.pad_token is None:
        special_tokens_dict["pad_token"] = DEFAULT_PAD_TOKEN  # 32000
    if tokenizer.eos_token is None:
        special_tokens_dict["eos_token"] = DEFAULT_EOS_TOKEN  # 2
    if tokenizer.bos_token is None:
        special_tokens_dict["bos_token"] = DEFAULT_BOS_TOKEN  # 1
    if tokenizer.unk_token is None:
        special_tokens_dict["unk_token"] = DEFAULT_UNK_TOKEN

    tokenizer.add_special_tokens(special_tokens_dict)

    tokenizer.add_tokens(
        [
            B_CODE,  # 32001
            E_CODE,  # 32002
            B_RESULT,  # 32003
            E_RESULT,  # 32004
            B_INST,
            E_INST,
            B_SYS,
            E_SYS,  # 32008
        ],
        special_tokens=True,
    )

    # build model
    if peft:
        model = LlamaForCausalLM.from_pretrained(
            hf_model_path,
            load_in_8bit=True,
            device_map="auto",
            ignore_mismatched_sizes=True,
            torch_dtype=torch.float16,
        )
    else:
        # for llama
        # model = LlamaForCausalLM.from_pretrained(
        #    hf_model_path, ignore_mismatched_sizes=True
        # )

        # for codellama
        from codellama_wrapper import CodeLlamaForCausalLM

        model = CodeLlamaForCausalLM.from_pretrained(hf_model_path)

    model.resize_token_embeddings(len(tokenizer))

    return {"tokenizer": tokenizer, "model": model}


def train():
    parser = transformers.HfArgumentParser(
        (ModelArguments, DataArguments, TrainingArguments)
    )
    model_args, data_args, training_args = parser.parse_args_into_dataclasses()

    model_dict = build_model_from_hf_path(
        hf_model_path=model_args.model_name_or_path, peft=model_args.peft
    )

    model, tokenizer = model_dict["model"], model_dict["tokenizer"]
    # peft setting
    model.train()
    if model_args.peft:
        model, lora_config = create_peft_config(model)

    # make dataset
    data_module = make_supervised_data_module(tokenizer=tokenizer, data_args=data_args)
    trainer = Trainer(
        model=model, tokenizer=tokenizer, args=training_args, **data_module
    )

    # train
    trainer.train()
    trainer.save_state()
    trainer.save_model(output_dir=training_args.output_dir)


if __name__ == "__main__":
    train()