-
Notifications
You must be signed in to change notification settings - Fork 0
/
model_infer.py
301 lines (282 loc) · 11.1 KB
/
model_infer.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
import argparse
import json
import logging
import os.path
from pathlib import Path
import torch
from tqdm import tqdm
from BaseNews import BaseNews
from vllm import LLM, SamplingParams
from transformers import (
AutoModelForCausalLM,
AutoTokenizer,
PreTrainedModel,
PreTrainedTokenizerBase,
BloomForCausalLM,
AutoModel,
LlamaForCausalLM
)
XINHUA_OBJECT = "xinhua_object"
XINHUA_SUBJECT = "xinhua_subject"
SAFE_OBJECT = "safe_object"
SAFE_SUBJECT = "safe_subject"
class SafeNews(BaseNews):
r"""
:param model_path:str model path
:param model_name:str model name
:param model_type:str model type
:param output_dir:str output dir of result
:param data_path:str dataset path
:param temperature:float
:param top_k:float
:param top_p:float
:param vllm:bool Whether to use vLLM acceleration
"""
def __init__(self,
model_path,
model_name,
model_type,
output_dir,
data_path,
temperature: float,
top_k: int,
top_p: float,
vllm: bool,
tensor_parallel_size,
max_num_batched_tokens):
super().__init__()
assert os.path.exists(model_path), "Model does not exist."
self.model_path = model_path
self.model_name = model_name
self.model_type = model_type
self.output_dir = output_dir
self.temperature = temperature
self.top_k = top_k
self.top_p = top_p
self.vllm = vllm
self.object_result = []
self.subject_result = []
self.xinhua_object_data = []
self.xinhua_subject_data = []
self.safe_object_data = []
self.safe_subject_data = []
self.is_load_model = self.model_type not in self.not_load_model_types
if tensor_parallel_size is None:
self.tensor_parallel_size = 8
else:
self.tensor_parallel_size = min(8, tensor_parallel_size)
if max_num_batched_tokens is None:
self.max_num_batched_tokens = 8192
else:
self.max_num_batched_tokens = max_num_batched_tokens
self.all_data = {
SAFE_OBJECT: self.safe_object_data,
SAFE_SUBJECT: self.safe_subject_data,
XINHUA_OBJECT: self.xinhua_object_data,
XINHUA_SUBJECT: self.xinhua_subject_data
}
# Initialize model.
if self.is_load_model:
if vllm:
self.llm, self.sampling_params = self.getModel()
else:
self.model, self.tokenizer = self.getModel()
# Initialize dataset.
if data_path is not None:
assert os.path.exists(data_path), "Dataset does not exist."
with open(data_path, "r", encoding='utf-8') as f:
self.eval_data = json.load(f)
self.processData()
def getModel(self):
r"""
This function returns an instance of the model used for evaluation.
If vllm is used, return llm, sampling_params.
Otherwise, return the original model, tokenizer.
:return:
"""
if self.vllm:
if self.model_type.lower() == "self-70b":
tokenizer = AutoTokenizer.from_pretrained(self.model_path)
added_tokens = {'cls_token': '<CLS>', 'sep_token': '<SEP>', 'additional_special_tokens': ['<EOD>'],
'mask_token': '<MASK>', 'pad_token': '<PAD>'}
tokenizer.add_special_tokens(added_tokens)
llm = LLM(model=self.model_path, tokenizer=self.model_path, tensor_parallel_size=self.tensor_parallel_size)
llm.set_tokenizer(tokenizer)
sampling_params = SamplingParams(temperature=self.temperature, top_p=0.7, max_tokens=100,
stop_token_ids=[2, 60301])
else:
llm = LLM(model=self.model_path, trust_remote_code=True,
max_num_batched_tokens=self.max_num_batched_tokens,
tokenizer_mode='auto', tensor_parallel_size=self.tensor_parallel_size)
sampling_params = SamplingParams(max_tokens=8192, temperature=self.temperature, top_k=20, top_p=0.7)
return llm, sampling_params
model = AutoModelForCausalLM.from_pretrained(
self.model_path,
trust_remote_code=True,
device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(
self.model_path,
trust_remote_code=True
)
return model, tokenizer
def processData(self):
for t in self.eval_data:
assert t['type'] in self.type_to_dict.keys(), f"Unknown type exists in the data: {t['type']}"
if self.type_to_dict[t['type']] == SAFE_OBJECT:
self.safe_object_data.append(t)
elif self.type_to_dict[t['type']] == SAFE_SUBJECT:
self.safe_subject_data.append(t)
elif self.type_to_dict[t['type']] == XINHUA_OBJECT:
self.xinhua_object_data.append(t)
elif self.type_to_dict[t['type']] == XINHUA_SUBJECT:
self.xinhua_subject_data.append(t)
def eval(self):
r"""
:return:
"""
if self.is_load_model:
if self.vllm:
self.inferenceByVllm()
else:
self.inference()
self.saveResult()
def inference(self):
if self.model_type == "qwen":
gen_kwargs = {
"do_sample": "True",
"eos_token_id": [
151645
],
"max_new_tokens": 2048,
"pad_token_id": 151643,
"repetition_penalty": 1.2,
"temperature": 0.5,
"top_k": 40,
"top_p": 0.7
}
else:
gen_kwargs = {
"temperature": self.temperature,
"top_k": self.top_k,
"top_p": self.top_p,
"repetition_penalty": 1.0,
"max_new_tokens": 4096,
}
with tqdm(total=len(self.eval_data)) as pbar:
pbar.set_description('inference:')
for key, value in self.all_data.items():
template = self.getTemplate(self.model_type, key)
for i in value:
prompt = template.format_map(i)
i["prompt"] = prompt
if self.model_type == "chatglm":
self.model = self.model.eval()
output, history = self.model.chat(self.tokenizer, prompt, history=[])
elif self.model_type == "aquilaChat":
from aquila_predict import predict
output = predict(self.model, prompt, tokenizer=self.tokenizer, max_gen_len=2048, top_p=0.9,
seed=123, topk=15, temperature=1.0, sft=True,
model_name="AquilaChat2-34B-16K")
elif self.model_type == "xverse":
self.model = self.model.eval()
history = [{"role": "user", "content": prompt}]
output = self.model.chat(self.tokenizer, history)
else:
inputs = self.tokenizer.encode(prompt, return_tensors="pt").cuda()
outputs = self.model.generate(inputs, **gen_kwargs)[0]
output = self.tokenizer.decode(outputs[len(inputs[0]) - len(outputs):], skip_special_tokens=True)
i['output'] = output
pbar.update(1)
def inferenceByVllm(self):
safe_subject_prompts = []
safe_object_prompts = []
xinhua_object_prompts = []
xinhua_subject_prompts = []
all_prompts = {
SAFE_OBJECT: safe_object_prompts,
SAFE_SUBJECT: safe_subject_prompts,
XINHUA_OBJECT: xinhua_object_prompts,
XINHUA_SUBJECT: xinhua_subject_prompts
}
logging.info("Prepare data.")
for key, value in all_prompts.items():
template = self.getTemplate(self.model_type, key)
for i in self.all_data[key]:
prompt = template.format_map(i)
i["prompt"] = prompt
value.append(prompt)
logging.info("Data preparation is complete.")
logging.info("Model generation begins.")
for key, value in all_prompts.items():
outputs = self.llm.generate(value, self.sampling_params)
cnt = 0
for i in self.all_data[key]:
i['output'] = outputs[cnt].outputs[0].text
cnt += 1
def saveResult(self):
if self.output_dir is not None:
output_dir = self.output_dir
else:
current_path = Path(__file__).resolve().parent
# Get the root directory of the current script's project.
root_path = current_path.parent
output_dir = os.path.join(root_path, f"output/{self.model_name}")
if not os.path.exists(output_dir):
os.makedirs(output_dir)
save_path = os.path.join(output_dir, f"{self.model_name}_output.json")
with open(save_path, "w", encoding="utf-8") as f:
json.dump(self.all_data, f, indent=2, ensure_ascii=False)
def parse_argument():
parser = argparse.ArgumentParser()
parser.add_argument(
"--model_name", type=str, help="model name"
)
parser.add_argument(
"--model_path", type=str, help="model path"
)
parser.add_argument(
"--model_type", type=str, default="default", help="model type"
)
parser.add_argument(
"--output_dir", type=str, default=None, help="output directory"
)
parser.add_argument(
"--data_path", type=str, default=None, help="data directory"
)
parser.add_argument(
"--temperature", type=float, default=0.1, help="temperature"
)
parser.add_argument(
"--top_k", type=int, default=20, help="top_k"
)
parser.add_argument(
"--top_p", type=float, default=0.7, help="top_p"
)
parser.add_argument(
"--vllm", action="store_true", help="vllm"
)
parser.add_argument(
"--tensor_parallel_size", type=int, default=8, help="tensor_parallel_size"
)
parser.add_argument(
"--max_num_batched_tokens", type=int, required=False, help="max_num_batched_tokens"
)
return parser.parse_args()
def main():
args = parse_argument()
safe = SafeNews(model_path=args.model_path,
model_name=args.model_name,
model_type=args.model_type,
output_dir=args.output_dir,
data_path=args.data_path,
temperature=args.temperature,
top_k=args.top_k,
top_p=args.top_p,
vllm=args.vllm,
tensor_parallel_size=args.tensor_parallel_size,
max_num_batched_tokens=args.max_num_batched_tokens)
print(args)
safe.eval()
if __name__ == '__main__':
main()