-
Notifications
You must be signed in to change notification settings - Fork 4
/
inference.py
138 lines (112 loc) · 3.9 KB
/
inference.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
import os
import sys
import editdistance
import torch
import transformers
from peft import PeftModel
from transformers import GenerationConfig, LlamaForCausalLM, LlamaTokenizer
import re
from utils.callbacks import Iteratorize, Stream
from utils.prompter import Prompter
from argparse import ArgumentParser
import json
import logging
if torch.cuda.is_available():
device = "cuda"
else:
device = "cpu"
try:
if torch.backends.mps.is_available():
device = "mps"
except: # noqa: E722
pass
load_8bit = True
def calculate_wer(pre, ref):
wer_score = editdistance.eval(pre, ref) / len(ref)
return wer_score
if __name__ == '__main__':
parser = ArgumentParser()
parser.add_argument("--ckpt_path", type=str, required=True)
parser.add_argument("--test_data_path", type=str, required=True, help='<your_test_data>.json')
args = parser.parse_args()
prompter = Prompter("alpaca")
tokenizer = LlamaTokenizer.from_pretrained('yahma/llama-13b-hf')
model = LlamaForCausalLM.from_pretrained(
'yahma/llama-13b-hf',
load_in_8bit=load_8bit,
torch_dtype=torch.float16,
device_map='auto',
)
model = PeftModel.from_pretrained(
model,
args.ckpt_path,
torch_dtype=torch.float16,
device_map={'': 0},
)
model.config.pad_token_id = 0
model.config.bos_token_id = 1
model.config.eos_token_id = 2
if not load_8bit:
model.half()
model.eval()
if torch.__version__ >= "2" and sys.platform != "win32":
model = torch.compile(model)
def alpaca_infer(input1, input2=None):
if input2 is not None:
prompt = prompter.generate_prompt(input=input1, input2=input2)
else:
prompt = prompter.generate_prompt(input=input1)
inputs = tokenizer(prompt, return_tensors="pt")
input_ids = inputs["input_ids"].to(device)
generation_config = GenerationConfig(
temperature=0.1,
top_p=0.75,
top_k=40,
num_beams=4,
max_new_tokens=256,
stream_output=False,
prompter=prompter,
model=model,
)
with torch.no_grad():
generation_output = model.generate(
input_ids=input_ids,
generation_config=generation_config,
return_dict_in_generate=True,
output_scores=True,
max_new_tokens=256,
)
s = generation_output.sequences[0]
output = tokenizer.decode(s)
return prompter.get_response(output)
with open(args.test_data_path, 'r') as f:
test_data = json.load(f)
before = 0
after = 0
ignore = 0
for i in range(len(test_data)):
best_hypo = test_data[i]['input1']
input2 = test_data[i]['input2']
ground_truth = test_data[i]['output']
prediction = alpaca_infer(input1=best_hypo, input2=input2)
prediction = re.sub('</s>', '', prediction).lower()
prediction = re.sub(r'[^\w\s]|[\d]', '', prediction)
best_hypo = re.sub(r'[^\w\s]|[\d]', '', best_hypo)
ground_truth = re.sub(r'[^\w\s]|[\d]', '', ground_truth)
prediction = re.sub(r'\n+.+', '', prediction)
try:
wer_prediction = calculate_wer(prediction.split(), ground_truth.split())
wer_best_hypo = calculate_wer(best_hypo.split(), ground_truth.split())
except Exception:
ignore += 1
continue
before = before + wer_best_hypo
after = after + wer_prediction
if wer_best_hypo != wer_prediction:
print('before:::', best_hypo)
print('after :::', prediction)
print('answer:::', ground_truth)
print('before score', wer_best_hypo)
print('after score', wer_prediction)
print('before', before / (len(test_data) - ignore))
print('after', after / (len(test_data) - ignore))