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
| import json
from shutil import copyfile
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
from settings import *
def gen_prompt(instruction, input=None, template=PROMPT_TEMPLATE):
"""Generate prompt before feeding inputs into LLM"""
if input:
instruction = instruction + '\n' + input
return template.format_map({'instruction': instruction})
class ModelInterface:
def __init__(self):
self.args = None
self.generation_config = None
self.model_path = None
self.model = None
self.tokenizer = None
self.device = None
self.setup()
def setup(self):
self.args = parse_args()
self.device = apply_settings(args=self.args)
self.model_path = model_path[self.args.model_name]
self.generation_config = GenerationConfig(
temperature=self.args.temperature,
repetition_penalty=self.args.repetition_penalty,
do_sample=self.args.do_sample,
num_beams=self.args.num_beams,
top_p=self.args.top_p,
top_k=self.args.top_k,
max_new_tokens=self.args.max_new_tokens,
)
self.load_model()
def load_model(self):
tokenizer = AutoTokenizer.from_pretrained(self.model_path)
base_model = AutoModelForCausalLM.from_pretrained(
self.model_path,
device_map='auto',
load_in_4bit=True
)
model_vocab_size = base_model.get_input_embeddings().weight.size(0)
tokenzier_vocab_size = len(tokenizer)
print(f">>> Model name: {self.args.model_name}")
print(f">>> Vocab of the base model: {model_vocab_size}")
print(f">>> Vocab of the tokenizer: {tokenzier_vocab_size}")
if model_vocab_size!=tokenzier_vocab_size:
assert tokenzier_vocab_size > model_vocab_size
print(">>> Resize model embeddings to fit tokenizer")
base_model.resize_token_embeddings(tokenzier_vocab_size)
if self.args.lora_model is not None:
print(">>> Loading peft model")
load_type = torch.float16
model = PeftModel.from_pretrained(
base_model,
self.args.lora_model,
torch_dtype=load_type,
device_map='auto',)
else:
model = base_model
if self.device==torch.device('cpu'):
model.float()
model.eval()
print(">>> Finish loading model and tokenizer")
self.model, self.tokenizer = model, tokenizer
def run_iter(self):
"""Run LLM iteratively"""
print(">>> Running LLM iteratively")
with torch.no_grad():
while True:
raw_input_text = input(">>> Input: ")
if len(raw_input_text.strip())==0:
break
if self.args.with_prompt:
input_text = gen_prompt(instruction=raw_input_text)
else:
input_text = raw_input_text
model_inputs = self.tokenizer(input_text, return_tensors="pt")
generated_ids = self.model.generate(
input_ids = model_inputs["input_ids"].to(self.device),
attention_mask = model_inputs['attention_mask'].to(self.device),
eos_token_id=self.tokenizer.eos_token_id,
pad_token_id=self.tokenizer.pad_token_id,
generation_config=self.generation_config
)[0]
output = self.tokenizer.decode(generated_ids, skip_special_tokens=True)
if self.args.with_prompt:
response = output.split("### Response:")[1].strip()
else:
response = output
print(">>> Response: ", response, "\n")
def run_batch(self, data, file_out):
print(">>> Feeding data to LLM in a batch")
results = []
with torch.no_grad():
for index, example in enumerate(data):
if self.args.with_prompt is True:
input_text = gen_prompt(instruction=example)
else:
input_text = example
inputs = self.tokenizer(input_text,return_tensors="pt")
generated_ids = self.model.generate(
input_ids = inputs["input_ids"].to(self.device),
attention_mask = inputs['attention_mask'].to(self.device),
eos_token_id=2,
pad_token_id=2,
generation_config=self.generation_config
)[0]
output = self.tokenizer.decode(generated_ids,skip_special_tokens=True)
if self.args.with_prompt:
response = output.split("### Response:")[1].strip()
else:
response = output
print(f"======={index}=======")
print(f"Input: \n{example}\n")
print(f"Output: \n{response}\n")
results.append({"Input":input_text, "Output":response})
if file_out:
if os.path.exists(file_out):
copyfile(file_out, file_out+".bak")
with open(file_out, 'w') as f:
json.dump(results, f, ensure_ascii=False, indent=2)
else:
return results
if __name__ == '__main__':
mi = ModelInterface()
mi.setup()
data = [
"Please tell me a joke.",
"Please ...",
]
file_out = "test.json"
mi.run_batch(data, file_out)
|
