def setup_class(self):
self.use_gpu = torch.cuda.is_available()
self.test_dir = Path(tempfile.mkdtemp())
self.base_tokenizer = CTRLTokenizer.from_pretrained('ctrl', do_lower_case=True,
cache_dir=self.test_dir)
self.rust_tokenizer = PyCtrlTokenizer(
get_from_cache(self.base_tokenizer.pretrained_vocab_files_map['vocab_file']['ctrl']),
get_from_cache(self.base_tokenizer.pretrained_vocab_files_map['merges_file']['ctrl']),
do_lower_case=True
)
self.model = CTRLModel.from_pretrained('ctrl',
output_attentions=False).eval()
if self.use_gpu:
self.model.cuda()
self.sentence_list = ['For instance, on the planet Earth, man had always assumed that he was more intelligent '
'than dolphins because he had achieved so much—the wheel, New York, wars and so on—whilst'
' all the dolphins had ever done was muck about in the water having a good time. But '
'conversely, the dolphins had always believed that they were far more intelligent than '
'man—for precisely the same reasons.'] * 1
# Pre-allocate GPU memory
tokens_list = [self.base_tokenizer.tokenize(sentence) for sentence in self.sentence_list]
features = [self.base_tokenizer.convert_tokens_to_ids(tokens) for tokens in tokens_list]
features = [self.base_tokenizer.prepare_for_model(input, None, add_special_tokens=True, max_length=128) for
input
in features]
all_input_ids = torch.tensor([f['input_ids'] for f in features], dtype=torch.long)
if self.use_gpu:
all_input_ids = all_input_ids.cuda()
with torch.no_grad():
_ = self.model(all_input_ids)[0].cpu().numpy()
def test_TFCTRL(self):
from transformers import CTRLTokenizer, TFCTRLModel
tokenizer = CTRLTokenizer.from_pretrained('ctrl')
model = TFCTRLModel.from_pretrained('ctrl')
input_dict = tokenizer("Hello, my dog is cute", return_tensors="tf")
spec, input_dict = self.spec_and_pad(input_dict)
outputs = ["last_hidden_state"]
self.run_test(model, input_dict, input_signature=spec, outputs=outputs, large=True)
def setup_class(self):
self.processor = Sst2Processor()
self.test_dir = Path(tempfile.mkdtemp())
sst2_url = 'https://firebasestorage.googleapis.com/v0/b/mtl-sentence-representations.appspot.com/o/data%2FSST-2.zip?alt=media&token=aabc5f6b-e466-44a2-b9b4-cf6337f84ac8'
contents = requests.get(sst2_url)
(self.test_dir / 'SST-2.zip').open('wb').write(contents.content)
with ZipFile(self.test_dir / 'SST-2.zip', 'r') as zipObj:
zipObj.extractall(self.test_dir)
self.examples = self.processor.get_train_examples(self.test_dir / 'SST-2')
self.base_tokenizer = CTRLTokenizer.from_pretrained('ctrl',
do_lower_case=False,
cache_dir=self.test_dir)
self.rust_tokenizer = PyCtrlTokenizer(
get_from_cache(self.base_tokenizer.pretrained_vocab_files_map['vocab_file']['ctrl']),
get_from_cache(self.base_tokenizer.pretrained_vocab_files_map['merges_file']['ctrl']),
do_lower_case=False)
def test_tokenization_ctrl(self):
# Given
self.base_tokenizer = CTRLTokenizer.from_pretrained(
'ctrl', do_lower_case=True, cache_dir=self.test_dir)
self.rust_tokenizer = PyCtrlTokenizer(
get_from_cache(
self.base_tokenizer.pretrained_vocab_files_map['vocab_file']
['ctrl']),
get_from_cache(
self.base_tokenizer.pretrained_vocab_files_map['merges_file']
['ctrl']),
do_lower_case=True)
output_baseline = []
for example in self.examples:
output_baseline.append(
self.base_tokenizer.encode_plus(
example.text_a,
add_special_tokens=True,
return_overflowing_tokens=True,
return_special_tokens_mask=True,
max_length=128))
# When
output_rust = self.rust_tokenizer.encode_list(
[example.text_a for example in self.examples],
max_len=128,
truncation_strategy='longest_first',
stride=0)
# Then
for idx, (rust,
baseline) in enumerate(zip(output_rust, output_baseline)):
assert rust.token_ids == baseline[
'input_ids'], f'Difference in tokenization for {self.rust_tokenizer.__class__}: \n ' \
f'Sentence a: {self.examples[idx].text_a} \n' \
f'Sentence b: {self.examples[idx].text_b} \n' \
f'Token mismatch: {self.get_token_diff(rust.token_ids, baseline["input_ids"])} \n' \
f'Rust: {rust.token_ids} \n' \
f'Python {baseline["input_ids"]}'
assert (rust.segment_ids == baseline['token_type_ids'])
assert (
rust.special_tokens_mask == baseline['special_tokens_mask'])
def setup_base_tokenizer(self):
self.base_tokenizer = CTRLTokenizer.from_pretrained('ctrl', do_lower_case=True,
cache_dir=self.test_dir)
import torch
from transformers import CTRLTokenizer, CTRLModel
tokenizer = CTRLTokenizer.from_pretrained('ctrl')
model = CTRLModel.from_pretrained('ctrl')
input_ids =torch.tensor(tokenizer.encode("Links Hello, my dog is cute",add_special_tokens=True)).unsqueeze(0) # Batch size 1 #因为只有一个句子所以需要unsqueeze去掉一层.
outputs = model(input_ids)# 输出的第一个是结果,第二个是cache没啥用.
print("Warning: Please set args.num_gens higher, unable to make equal spread across p values and prompt lengths")
exit(0)
if args.control_code and args.control_code not in CONTROL_CODES and args.control_code != "Politics":
print("Warning: Invalid control code selection, please select a valid control code")
exit(0)
# Download the sampling file from the Google Cloud bucket
filename = 'prompts-{}.json'.format('-'.join([args.dataset, args.split]))
file_url = 'gs://roft_datasets/prompts/' + args.dataset + '/' + filename
local_file_path = './' + args.dataset + '/' + filename
command = "gsutil cp {0} {1}".format(file_url, local_file_path)
process = subprocess.Popen(command.split(), stdout=subprocess.PIPE)
# Initialize the tokenizer and the model
tokenizer = CTRLTokenizer.from_pretrained(args.model_name)
model = TFCTRLLMHeadModel.from_pretrained(args.model_name, return_dict=True)
# Calculate the 99 percentile length of a prompt with MIN_NUM_SENTS sentences in it
with open(local_file_path, 'r') as f:
data = json.load(f)
tokenized_prompts = []
for prompt in data['prompts']:
inputs = tokenizer(' '.join([args.control_code] + prompt[:MIN_NUM_SENTS]), return_tensors="tf")
tokenized_prompts.append(float(len(inputs['input_ids'][0])) / float(MIN_NUM_SENTS))
sorted_lens = sorted(tokenized_prompts)
ninety_percentile_sent_len = sorted_lens[int(0.90*len(sorted_lens))]
nlp = spacy.load("en")
set_seed(RANDOM_SEED)