def fun(S, path_config, gpu, nsamples):
model, tokenizer, config, device = getModel(path_config=path_config,
gpu=gpu)
path_target = 'data/new.txt'
#if not os.path.exists(path_target):
#os.mkdir(path_target)
N = 0
n = 0
for i in range(len(S)):
if i % 10 == 0:
print(
"proceed {} poem (total {}), get {} prefix and generate {} poems"
.format(i, len(S), n, N))
sents = peomSplit(S[i])
for s in sents:
n += 1
if '□' in s:
continue
r = generating(s, model, config, tokenizer, device, nsamples, gpu)
N += len(r)
if len(r) == 0:
continue
r = [s + '\t' + rr[len(s):] for rr in r]
with open(path_target, 'a+') as f:
f.write('\n'.join(r) + '\n')
f.close()
def main(path_data, mode, path_config, path_target, topk, temp):
ii = int(mode)
model, tokenizer, config, device = gpt_gen.getModel(
path_config=path_config)
config['topk'] = topk
config['temperature'] = temp
with open(path_data, 'r') as f:
s = f.read().strip().split('\n')
D = []
t0 = time.time()
for data in s:
result = []
for _ in range(1):
if ii == 1:
r0 = gpt_gen.generating_poem(app,
data,
model,
config,
tokenizer,
device,
quick,
num0[ii],
batchGenerating=batchGenerating)
else:
r0 = gpt_gen.generating(app,
data,
model,
config,
tokenizer,
device,
ConfigPredict,
quick=quick,
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))
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]
for gpu in gpus:
m0,t0,c0,d0 = gpt_gen.getModel(path_config=path_configs[ii],gpu=gpu)
c0['repetition_penalty'] = ConfigPredict.repetition_penalty[ii]
c0['temperature'] = ConfigPredict.temperature[ii]
c0['length'] = ConfigPredict.length[ii]
M0.append(m0)
T0.append(t0)
C0.append(c0)
D0.append(d0)
model.append(M0)
tokenizer.append(T0)
config.append(C0)
device.append(D0)
ModelIndex.append([kk for kk in range(len(gpus))])
GPUs.append(gpus)
D_simi = json.load(open(path_simi,'r',encoding='utf-8'))
D_next = json.load(open(path_next,'r',encoding='utf-8'))
num0 = [c.predict_nums for c in ConfigPredict]
tags = [c.tags for c in ConfigPredict]
rmHFW = [c.rmHFW for c in ConfigPredict]
'''
maxNext = ConfigPredict.maxNext_JLX
path_next = ConfigPredict.path_JLX_next
path_simi = ConfigPredict.path_JLX_simi
D_simi = json.load(open(path_simi,'r',encoding='utf-8'))
D_next = json.load(open(path_next,'r',encoding='utf-8'))
D_simi = {k:json.loads(D_simi[k]) for k in D_simi}
D_next = {k:json.loads(D_next[k]) for k in D_next}
'''
D_simi,D_next,maxNext=[],[],[]
model,tokenizer,config,device = [], [], [], []
for ii in range(len(path_configs)):
m0,t0,c0,d0 = gpt_gen.getModel(path_config=path_configs[ii],gpu=ConfigPredict[ii].gpus)
c0['repetition_penalty'] = ConfigPredict[ii].repetition_penalty
c0['temperature'] = ConfigPredict[ii].temperature
c0['length'] = ConfigPredict[ii].length
model.append(m0)
tokenizer.append(t0)
config.append(c0)
device.append(d0)
@app.route('/api/gen', methods=['POST'])
def test2():
r = request.json
#print(type(r))
#print(request.json)
#r = '{"input": "们"}'
#r = json.loads(r)
data = r["input"]
port = 7000
from Config_gou import config_predict
gpus = '0,1,2,3'
style = 'gou'
ConfigPredict = config_predict(gpus=gpus)
batchGenerating = ConfigPredict.batchGenerating
path_configs = ConfigPredict.model_configs
num0 = ConfigPredict.predict_nums
tags = ConfigPredict.tags
rmHFW = ConfigPredict.rmHFW
gpus = ConfigPredict.gpus.split(',')
Model = []
for gpu in gpus:
model, tokenizer, config, device = gpt_gen.getModel(
path_config=path_configs, gpu=gpu)
Model.append((model, gpu))
@app.route('/', methods=['POST'])
def test1():
r = request.json
data = r["input"]
quick = False
if "quick" in r:
print("quick pattern")
if r["quick"] == "True":
quick = True
try:
random.shuffle(Model)
model, gpu = Model[0][0], Model[0][1]