def generating_poem(app,prefix,model,config,tokenizer,device,quick=False,num=5,batchGenerating=False,gpu='0',onlyMax=False,fast_pattern=False):
torch.cuda.set_device(int(gpu))
if len(prefix)>10:
return []
#print("start:", prefix)
global a
a = app
n_ctx = model.config.n_ctx
length = config['length']
nsamples = num
batch_size = config['batch_size']
temperature = config['temperature']
topk = config['topk']
topp = config['topp']
quick_pattern = quick
repetition_penalty = config['repetition_penalty']
if length == -1:
length = model.config.n_ctx
#print('generating-begin for %s'%prefix)
raw_text = prefix[0]+prefix
context_tokens = tokenizer.convert_tokens_to_ids(tokenizer.tokenize(raw_text))
if batchGenerating:
if onlyMax:
outs = sample_sequence_batch_max(model, context_tokens, length, n_ctx, tokenizer, nsamples=2,
temperature=temperature, top_k=topk,
top_p=topp, repitition_penalty=repetition_penalty,
device=device)
else:
if fast_pattern:
outs = fast_sample_sequence_batch(model, context_tokens, length, nsamples=nsamples,
temperature=temperature, top_k=topk,
repitition_penalty=repetition_penalty, device=device)
else:
outs = sample_sequence_batch_opti(model, context_tokens, length, n_ctx, tokenizer, nsamples, temperature=temperature, top_k=topk,
top_p=topp, repitition_penalty=repetition_penalty,
device=device)
S = []
for out in outs:
tmptext = untokenization_poem(out, tokenizer, config)
poem = poemFilter1(tmptext[1:])
if poem:
S.append(poem)
else:
S = []
for _ in range(nsamples):
out = generate(
n_ctx=n_ctx,
model=model,
context=context_tokens,
length=length,
is_fast_pattern=fast_pattern, tokenizer=tokenizer,
temperature=temperature, top_k=topk, top_p=topp, repitition_penalty=repetition_penalty, device=device
)
tmptext = untokenization_poem(out, tokenizer, config)
poem = poemFilter1(tmptext[1:])
if poem:
S.append(poem)
S = dropDuplicateContent(S)
return S
def getpoem(len_sent, nb_sents):
#len_sent = 7
#nb_sents = 8
if len(prefix) < nb_sents:
prefix0 = list(prefix) + [''] * (nb_sents - len(prefix) + 1)
else:
prefix0 = list(prefix)
raw_text = '[MASK]' + prefix0[0]
#raw_text = prefix0[0] + prefix0[0]
context = tokenizer.convert_tokens_to_ids(tokenizer.tokenize(raw_text))
contexts = [[c for c in context] for _ in range(nsamples)]
inputs = [[c for c in context] for _ in range(nsamples)]
num = nsamples
for ii in range(1, nb_sents + 1):
outs = fast_sample_sequence_batch_poemHead(
model,
contexts,
inputs,
length=len_sent + 1,
nsamples=num,
temperature=temperature,
top_k=topk,
repitition_penalty=repetition_penalty,
device=device)
S = [untokenization_poem(out, tokenizer, config) for out in outs]
if ii == nb_sents:
break
S = [
tmptext for tmptext in S
if len(tmptext) > ii * (len_sent + 1) - 1
]
if ii % 2 == 0:
S1 = [
tt[:ii * (len_sent + 1)] for tt in S
if tt[ii * (len_sent + 1) - 1] in punc_end
]
else:
S1 = [
tt[:ii * (len_sent + 1)] for tt in S
if tt[ii * (len_sent + 1) - 1] in punc_mid
]
raw_texts = ['[MASK]' + s + prefix0[ii] for s in S1]
num = len(raw_texts)
contexts = [
tokenizer.convert_tokens_to_ids(tokenizer.tokenize(raw))
for raw in raw_texts
]
inputs = contexts
if num == 0:
break
R = []
for s in S:
if s[-1] not in punc_end:
t = s + '。'
else:
t = s
poem = poemFilter1(t, prefix, config_predict.blackwords)
if poem:
R.append(poem)
return R
def generating_poem(app,
prefix,
model,
config,
tokenizer,
device,
config_predict,
quick=False,
num=5,
continue_writing=False,
removeHighFreqWords=False,
batchGenerating=False,
gpu='0',
onlyMax=False,
maxNb=20):
if len(prefix) == 0 or len(prefix) > model.config.n_ctx:
return []
if sum([_is_chinese_char(c) for c in prefix]) < len(prefix) * 0.75:
return []
if gpu:
torch.cuda.set_device(int(gpu))
device = "cuda" if torch.cuda.is_available() else "cpu"
else:
device = 'cpu'
punc = '.,?!;\t 。，？！；'
global a
a = app
fast_pattern = config_predict.fast_pattern
n_ctx = model.config.n_ctx
length = config['length']
nsamples = num
#maxNb = max(nsamples,maxNb)
maxNb = nsamples
temperature = config['temperature']
topk = config['topk']
topp = config['topp']
quick_pattern = quick
repetition_penalty = config['repetition_penalty']
if length == -1:
length = model.config.n_ctx
#raw_text = prefix[0] + prefix
raw_text = '[MASK]' + prefix
context_tokens = tokenizer.convert_tokens_to_ids(
tokenizer.tokenize(raw_text))
t0 = time.time()
outs = []
if batchGenerating:
S = []
if onlyMax:
outs = sample_sequence_batch_max(
model,
context_tokens,
length,
n_ctx,
tokenizer,
nsamples=2,
temperature=temperature,
top_k=topk,
top_p=topp,
repitition_penalty=repetition_penalty,
device=device)
else:
if fast_pattern:
outs = fast_sample_sequence_batch(
model,
context_tokens,
length,
nsamples=maxNb,
temperature=temperature,
top_k=topk,
repitition_penalty=repetition_penalty,
device=device)
else:
outs = sample_sequence_batch_opti(
model,
context_tokens,
length,
n_ctx,
tokenizer,
maxNb,
temperature=temperature,
top_k=topk,
top_p=topp,
repitition_penalty=repetition_penalty,
device=device)
else:
S = []
for _ in range(maxNb):
out = generate(n_ctx=n_ctx,
model=model,
context=context_tokens,
length=length,
is_fast_pattern=fast_pattern,
tokenizer=tokenizer,
is_quick=quick_pattern,
temperature=temperature,
top_k=topk,
top_p=topp,
repitition_penalty=repetition_penalty,
device=device)
tmptext = untokenization(out, config, tokenizer, punc,
continue_writing)
S.append(tmptext)
S = []
for out in outs:
tmptext = untokenization_poem(out, tokenizer, config)
poem = poemFilter1(tmptext, prefix, config_predict.blackwords)
if poem:
S.append(poem)
S = dropDuplicateContent(S)
S = S[:nsamples]
return S