-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
322 lines (245 loc) · 11.3 KB
/
train.py
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
import os
import bitsandbytes as bnb
from peft import LoraConfig , get_peft_model , prepare_model_for_kbit_training , AutoPeftModelForCausalLM
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, set_seed, Trainer, TrainingArguments, BitsAndBytesConfig, \
DataCollatorForLanguageModeling, Trainer, TrainingArguments
from datasets import load_dataset
from dataclasses import dataclass, field
from typing import Dict, Optional, Sequence
import transformers
import copy
from accelerate import Accelerator
from torch.utils.data import Dataset
from datasets import concatenate_datasets
from peft.tuners.lora import LoraLayer
import random
def format_output(st):
return f"""{st['chat']}"""
def format_input(st):
return st['system']
IGNORE_INDEX = -100
DEFAULT_PAD_TOKEN = "<|pad|>"
DEFAULT_EOS_TOKEN = "<|endoftext|>"
DEFAULT_UNK_TOKEN = "<|unk|>"
@dataclass
class ModelArguments:
model_path: Optional[str] = field(
default="meta-llama/Llama-2-13b-hf")
@dataclass
class DataArguments:
data_path: str = field(default=None, metadata={
"help": "Path to the training data"})
num_examples: int = field(default=1, metadata={
"help": "num of examples"
})
@dataclass
class TrainingArguments(TrainingArguments):
cache_dir: Optional[str] = field(default=None)
optim: str = field(default="adamw_torch_fused")
model_max_length: int = field(
default=4096,
metadata={"help": "Maximum Sequence length"}
)
def smart_tokenizer_and_embedding_resize(special_tokens_dict: Dict, tokenizer: transformers.PreTrainedTokenizer , model: transformers.PreTrainedModel) -> Dict:
num_new_tokens = tokenizer.add_special_tokens(special_tokens_dict)
model.resize_token_embeddings(len(tokenizer))
if num_new_tokens > 0:
input_embeddings = model.get_input_embeddings().weight.data
output_embeddings = model.get_output_embeddings().weight.data
input_embeddings_avg = input_embeddings[:-num_new_tokens].mean(dim=0 , keepdim=True)
output_embeddings_avg = output_embeddings[:-num_new_tokens].mean(dim=0 , keepdim=True)
input_embeddings[-num_new_tokens:] = input_embeddings_avg
output_embeddings[-num_new_tokens:] = output_embeddings_avg
def _tokenize_fn(strings: Sequence[str] , tokenizer: transformers.PreTrainedTokenizer) -> Dict:
tokenized_list = [tokenizer(text , return_tensors="pt" , padding="longest" , max_length=tokenizer.model_max_length,
truncation=True
) for text in strings]
input_ids = labels = [tokenized.input_ids[0] for tokenized in tokenized_list]
#Get total input length by not including the padding tokens
input_id_lens = labels_lens = [tokenized.input_ids.ne(tokenizer.pad_token_id).sum().item() for tokenized in tokenized_list]
return dict(
input_ids=input_ids,
labels=labels,
input_id_lens=input_id_lens,
labels_lens=labels_lens
)
def preprocess(sources: Sequence[str] , target: Sequence[str] , tokenizer: transformers.PreTrainedTokenizer , dataset_type: str) -> Dict:
examples = [s + t for s,t in zip(sources , target)]
example_tokenized , source_tokenized = [_tokenize_fn(strings , tokenizer) for strings in (examples , sources)]
input_ids = example_tokenized["input_ids"]
lis = []
final_labels = []
labels = copy.deepcopy(input_ids)
# for i , label , source_len in zip(input_ids , labels , source_tokenized["input_id_lens"]):
# label[:source_len] = IGNORE_INDEX
# lis.append(i)
# final_labels.append(label)
c = 0
for ip , label in zip(input_ids , labels):
temp_label = label.tolist()
i = 0
while i < len(temp_label):
if temp_label[i] == 4816:
while i < len(temp_label):
if (temp_label[i] == 11123 or temp_label[i] ==16368):
break
i = i + 1
if i < len(temp_label):
temp_label[i] = -100
if i >= len(temp_label):
break
else:
temp_label[i] = -100
i = i + 1
final_labels.append(torch.tensor(temp_label))
lis.append(ip)
print("original dataset size" , len(input_ids))
print("final dataset size" , len(lis))
print("number of dropped records" , c)
return dict(input_ids=lis , labels=final_labels)
class SupervisedDataset(Dataset):
def __init__(self , tokenizer:transformers.PreTrainedTokenizer , main_dataset , dataset_type : str):
super(SupervisedDataset , self).__init__()
self.dataset = main_dataset
targets = []
sources = []
for i in self.dataset:
instruction = format_input(i)
output = format_output(i) + tokenizer.eos_token
sources.append(instruction)
targets.append(output)
data_dict = preprocess(sources=sources , target=targets , tokenizer=tokenizer , dataset_type=dataset_type)
self.input_ids = data_dict["input_ids"]
self.labels = data_dict["labels"]
print(f"length of {dataset_type}" , len(self.input_ids))
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):
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:
print("data path" , data_args.data_path)
main_dataset = load_dataset(data_args.data_path , split="train" , cache_dir = "./")
main_dataset = main_dataset.train_test_split(test_size=0.02, shuffle=True, seed=42)
print("train dataset" , main_dataset)
train_data = main_dataset['train']
eval_data = main_dataset['test']
train_dataset = SupervisedDataset(tokenizer = tokenizer , main_dataset=train_data , dataset_type="Train")
eval_dataset = SupervisedDataset(tokenizer = tokenizer , main_dataset=eval_data , dataset_type="Eval")
data_collator = DataCollatorForSuperVisedDataset(tokenizer=tokenizer)
return dict(train_dataset = train_dataset , eval_dataset = eval_dataset , data_collator = data_collator)
def create_bnb_config():
"""
bnb config to load the pre trained LM and
setting computation.
"""
bnb_config = BitsAndBytesConfig(
load_in_4bit=True,
bnb_4bit_use_double_quant=True,
bnb_4bit_quant_type="nf4",
bnb_4bit_compute_dtype=torch.bfloat16,
)
return bnb_config
def create_peft_config(modules):
"""
Create Parameter-Efficient Fine-Tuning config for the model
:param modules: Names of the modules to apply Lora to
"""
config = LoraConfig(
r=16, # dimension of the updated matrices
lora_alpha=64, # parameter for scaling
target_modules=modules,
lora_dropout=0.1, # dropout probability for layers
bias="none",
task_type="CAUSAL_LM",
)
return config
def find_all_linear_names(model, bits=4):
cls = bnb.nn.Linear4bit if bits == 4 else (bnb.nn.Linear8bitLt if bits == 8 else torch.nn.Linear)
lora_module_names = set()
for name, module in model.named_modules():
if isinstance(module, cls):
names = name.split('.')
lora_module_names.add(names[0] if len(names) == 1 else names[-1])
if 'lm_head' in lora_module_names: # needed for 16-bit
lora_module_names.remove('lm_head')
return list(lora_module_names)
def print_trainable_parameters(model, use_4bit=False):
"""
Prints the number of trainable parameters in the model.
"""
trainable_params = 0
all_param = 0
for _, param in model.named_parameters():
num_params = param.numel()
# if using DS Zero 3 and the weights are initialized empty
if num_params == 0 and hasattr(param, "ds_numel"):
num_params = param.ds_numel
all_param += num_params
if param.requires_grad:
trainable_params += num_params
if use_4bit:
trainable_params /= 2
print(
f"all params: {all_param:,d} || trainable params: {trainable_params:,d} || trainable%: {100 * trainable_params / all_param}"
)
def train():
parser = transformers.HfArgumentParser((ModelArguments , DataArguments , TrainingArguments))
model_args , data_args , training_args = parser.parse_args_into_dataclasses()
print("Model args" , model_args)
print("data_args" , data_args)
print("training_args" , training_args)
#create bnb config
bnb_config = create_bnb_config()
model = AutoModelForCausalLM.from_pretrained(
training_args.cache_dir,
quantization_config = bnb_config,
device_map={"": Accelerator().process_index})
model.config.pretraining_tp = 1
print("Model is" , model_args.model_path)
print("max length is" , training_args.model_max_length)
tokenizer = AutoTokenizer.from_pretrained(training_args.cache_dir , padding_side="right" , trust_remote_code=True , model_max_length = training_args.model_max_length)
model = prepare_model_for_kbit_training(model)
modules = find_all_linear_names(model , 4)
peft_config = create_peft_config(modules=modules)
model.enable_input_require_grads()
model = get_peft_model(model , peft_config)
print_trainable_parameters(model)
special_tokens_dict = dict()
if tokenizer.pad_token is None:
special_tokens_dict["pad_token"] = DEFAULT_PAD_TOKEN
if tokenizer.eos_token is None:
special_tokens_dict["eos_token"] = DEFAULT_EOS_TOKEN
if tokenizer.unk_token is None:
special_tokens_dict["unk_token"] = DEFAULT_UNK_TOKEN
print("special tokens dict" , special_tokens_dict)
#Resize Embeddings
smart_tokenizer_and_embedding_resize(special_tokens_dict=special_tokens_dict,
tokenizer=tokenizer,
model= model
)
#Process data
data_module = make_supervised_data_module(tokenizer=tokenizer , data_args = data_args)
#Printig One Example from the training data
print("train dataset Example" , data_module["train_dataset"][0])
print(tokenizer.decode(data_module["train_dataset"][0]["input_ids"]))
#Create Trainer
trainer = Trainer(model = model , tokenizer=tokenizer , args=training_args , **data_module)
trainer.train()
#trainer.save_state()
model.save_pretrained(training_args.output_dir)
#trainer.save_model(output_dir = training_args.output_dir)
if __name__ == "__main__":
train()