def SendStatement():
sentence = request.form['sentence']
response = {'success': True, 'processedSentence': None}
# try:
# body = request.get_json() or dict()
# except:
# body = dict()
# sentence = body.get('sentence')
# if not sentence:
# response['success'] = False
# LOGGER.error("Wrong key provided")
# # return jsonify(response)
# return render_template("public/response_sentence.html", page_body=response)
try:
processed_sentence = Monitors.Monitor_InputSentence(sentence)
processed_sentence = processSentence(processed_sentence)
except Exception as e:
print(e.stack)
response['success'] = False
LOGGER.error("Failed to process sentence")
# return jsonify(response)
return render_template("public/response_sentence.html",
page_body=response)
response['processedSentence'] = str(processed_sentence)
LOGGER.info("request successfully processed")
# return jsonify(response)
print(response)
return render_template("public/response_sentence.html",
page_body="Generated sentence is: {}".format(
response['processedSentence']))
def wrapper(*args, **kwargs):
authentication_token = request.form.get('AuthorizationToken')
if authentication_token == AUTHORIZATION_TOKEN_EXPECTED:
LOGGER.info("Successful auth")
return _function(*args, **kwargs)
else:
LOGGER.error(
"Failed auth with token {}".format(authentication_token))
return render_template(
"public/insert_sentence.html"
), requests.status_codes.codes['unauthorized']
def processSentence(sentence):
LOGGER.info("start processing sentence: {}".format(sentence))
try:
# output = subprocess.check_output('python3 src/interactive_conditional_samples.py --raw_text "{}"'.format(sentence),shell=True)
# response = output.decode('utf-8').split("\n")[-2]
# print(response)
# return response
PARAMS_DICT["model_name"] = 'beta2'
PARAMS_DICT["seed"] = None
PARAMS_DICT["nsamples"] = 1
PARAMS_DICT["batch_size"] = 1
PARAMS_DICT["length"] = None
PARAMS_DICT["temperature"] = 1
PARAMS_DICT["top_k"] = 40
PARAMS_DICT["top_p"] = 0.0
PARAMS_DICT["raw_text"] = sentence
NEW_PARAMS_DICT = Monitors.monitor_parameters(PARAMS_DICT)
assert NEW_PARAMS_DICT["nsamples"] % NEW_PARAMS_DICT["batch_size"] == 0
enc = encoder.get_encoder(NEW_PARAMS_DICT["model_name"])
hparams = model.default_hparams()
with open(
os.path.join('models', NEW_PARAMS_DICT["model_name"],
'hparams.json')) as f:
hparams.override_from_dict(json.load(f))
if NEW_PARAMS_DICT["length"] is None:
NEW_PARAMS_DICT["length"] = hparams.n_ctx // 2
elif NEW_PARAMS_DICT["length"] > hparams.n_ctx:
raise ValueError("Can't get samples longer than window size: %s" %
hparams.n_ctx)
with tf.Session(graph=tf.Graph()) as sess:
context = tf.placeholder(tf.int32,
[NEW_PARAMS_DICT["batch_size"], None])
np.random.seed(NEW_PARAMS_DICT["seed"])
tf.set_random_seed(NEW_PARAMS_DICT["seed"])
output = sample.sample_sequence(
hparams=hparams,
length=NEW_PARAMS_DICT["length"],
context=context,
batch_size=NEW_PARAMS_DICT["batch_size"],
temperature=NEW_PARAMS_DICT["temperature"],
top_k=NEW_PARAMS_DICT["top_k"],
top_p=NEW_PARAMS_DICT["top_p"])
saver = tf.train.Saver()
ckpt = tf.train.latest_checkpoint(
os.path.join('models', NEW_PARAMS_DICT["model_name"]))
saver.restore(sess, ckpt)
while True:
context_tokens = enc.encode(NEW_PARAMS_DICT["raw_text"])
generated = 0
for _ in range(NEW_PARAMS_DICT["nsamples"] //
NEW_PARAMS_DICT["batch_size"]):
out = sess.run(
output,
feed_dict={
context: [
context_tokens
for _ in range(NEW_PARAMS_DICT["batch_size"])
]
})[:, len(context_tokens):]
for i in range(NEW_PARAMS_DICT["batch_size"]):
generated += 1
text = enc.decode(out[i])
return text.split('<|endoftext|>')[0]
except:
traceback.print_exc()
exit()
def extract_dataset(tar_url, extract_path):
LOGGER.info("Extracting Dataset")
tar = tarfile.open(tar_url, 'r')
for item in tar:
tar.extract(item, extract_path)