def demo_resort():
path_data = 'data/test_text.txt'
path_source = 'result/test_text.json'
path_target = 'result/test_text-resort.json'
from Config import config_predict
import json
config = config_predict()
with open(path_data, 'r') as f:
Data = f.read().strip().split('\n')
with open(path_source, 'r') as f:
S = json.load(f)
D = []
for i in range(len(Data)):
prefix = Data[i]
outputs = S[i]['result']
s1 = [s[:-3] for s in outputs if '(文)' in s]
s2 = [s[:-5] for s in outputs if '(大白狗)' in s]
if len(s1) == 0:
r1 = []
else:
r1 = resort(prefix, s1, config)
if len(s2) == 0:
r2 = []
else:
r2 = resort(prefix, s2, config)
R = [r + '(文)' for r in r1]
R += [r + '(大白狗)' for r in r2]
d = {'input': prefix, 'outputs': R}
D.append(d)
with open(path_target, 'w') as f:
json.dump(D, f, ensure_ascii=False, indent=4)
def main(path_source, path_config, idx0, idx1, gpu, nsamples):
model, tokenizer, config, device = getModel(path_config=path_config, gpu=gpu)
from Config import config_predict
configpredict = config_predict()
with open(path_source, 'r') as f:
data = json.load(f)
S0 = data[idx0:idx1]
batch_size=100
i0 = 0
i1 = i0+batch_size
while i0<len(S0):
print('##########################%d-%d################'%(idx0+i0,idx0+min(i1,len(S0))))
S = S0[i0:i1]
fun(S, path_config, gpu, nsamples,model, tokenizer, config, device,configpredict)
i0 = i1
i1 = i1+batch_size
import gpt_gen_thread
import sys
import time
import logging
import torch
from Config import config_predict
from datetime import datetime
import GPUtil
#from gevent.pywsgi import WSGIServer #关键这个
app = Flask(__name__)
app.logger.setLevel(logging.INFO)
port = 5000
style = 0#0大白狗, 1散文
if len(sys.argv)>1:
port = int(sys.argv[1])
ConfigPredict = config_predict()
batchGenerating=ConfigPredict.batchGenerating
path_configs = ConfigPredict.model_configs
num0 = ConfigPredict.predict_nums
tags = ConfigPredict.tags
rmHFW = ConfigPredict.rmHFW
maxNext = ConfigPredict.maxNext_JLX
path_next = ConfigPredict.path_JLX_next
path_simi = ConfigPredict.path_JLX_simi
model,tokenizer,config,device,GPUs = [],[],[],[],[]
ModelIndex = []
for ii in range(len(path_configs)):
M0,T0,C0,D0 = [],[],[],[]
gpus = ConfigPredict.gpus[ii].split(',')
idx = path_configs[ii].index('config_')+len('config_')
key = path_configs[ii][idx:-5]
num=num0[ii],
removeHighFreqWords=rmHFW[ii],
batchGenerating=batchGenerating)
r0 = [rr + tags[ii] for rr in r0]
result.extend(r0)
d = {'input': data, 'outputs': result, 'num': len(result)}
D.append(d)
with open(path_target, 'w') as f:
json.dump(D, f, ensure_ascii=False, indent=4)
t1 = time.time()
print('predict time is {} for parameter topk={}'.format(t1 - t0, topk))
if __name__ == '__main__':
mode, path_config, data, path_target = sys.argv[1:5]
if len(sys.argv) > 6:
topk = int(sys.argv[5])
temp = float(sys.argv[6])
else:
topk = 8
temp = 1.0
ConfigPredict = config_predict(path_config)
batchGenerating = ConfigPredict.batchGenerating
path_configs = ConfigPredict.model_configs
num0 = ConfigPredict.predict_nums
tags = ConfigPredict.tags
rmHFW = ConfigPredict.rmHFW
maxNext = ConfigPredict.maxNext_JLX
path_next = ConfigPredict.path_JLX_next
path_simi = ConfigPredict.path_JLX_simi
main(data, mode, path_config, path_target, topk, temp)
import gpt_gen
import gpt_gen_thread
import sys
from datetime import datetime
import time
import logging
from Config import config_predict
path_source = sys.argv[1]
path_target = sys.argv[2]
if len(sys.argv) > 3:
path_config = sys.argv[3].split(',')
doPredict = [int(t) for t in sys.argv[4].split(',')]
gpus = sys.argv[5].split(',')
ConfigPredict = config_predict(model_config=path_config,
doPredict=doPredict,
gpus=gpus)
print('use input configs:%s' % '\n'.join(path_config))
else:
print('use default configs')
ConfigPredict = config_predict()
with open(path_source, 'r') as f:
Data = f.read().strip().split('\n')
batchGenerating = ConfigPredict.batchGenerating
path_configs = ConfigPredict.model_configs
num0 = ConfigPredict.predict_nums
tags = ConfigPredict.tags
rmHFW = ConfigPredict.rmHFW
maxNext = ConfigPredict.maxNext_JLX
path_next = ConfigPredict.path_JLX_next
import torch
import numpy as np
import gpt_gen
import gpt_gen_thread
import sys
from datetime import datetime
import time
import logging
from Config import config_predict
path_source = sys.argv[1]
path_target = sys.argv[2]
if len(sys.argv) > 3:
path_config = sys.argv[3].split(',')
doPredict = [int(t) for t in sys.argv[4].split(',')]
ConfigPredict = config_predict(model_config=path_config,
doPredict=doPredict)
print('use input configs:%s' % '\n'.join(path_config))
else:
print('use default configs')
ConfigPredict = config_predict()
with open(path_source, 'r') as f:
Data = f.read().strip().split('\n')
batchGenerating = ConfigPredict.batchGenerating
path_configs = ConfigPredict.model_configs
num0 = ConfigPredict.predict_nums
tags = ConfigPredict.tags
rmHFW = ConfigPredict.rmHFW
maxNext = ConfigPredict.maxNext_JLX
path_next = ConfigPredict.path_JLX_next
path_simi = ConfigPredict.path_JLX_simi