def generate_samples():
_data_c, _data_x = zip(*gen.next())
samples = fake_sample.eval(session=session,
feed_dict={
real_inputs_discrete_c: _data_c,
real_inputs_discrete_x: _data_x
})
decoded_samples = []
for i in xrange(len(samples)):
decoded_samples.append('Q: ' + Parser.decodeData(_data_c[i]))
decoded_samples.append('A: ' + Parser.decodeData(samples[i]))
decoded_samples.append('\n')
return decoded_samples
print 'Usage: python chatbot.py [Text pickle] [TRT Model]'
sys.exit(0)
with open(sys.argv[1]) as f:
wmap, iwmap = pickle.load(f)
Parser.runtimeLoad(wmap, iwmap)
print '[ChatBot] load word map from ' + sys.argv[1]
engine = tensorNet.createTrtFromUFF(sys.argv[2])
tensorNet.prepareBuffer(engine)
print '[ChatBot] create tensorrt engine from ' + sys.argv[2]
while True:
b = raw_input('\n\n\x1b[1;105;97m' +
'Please write your question(q for quite):' + '\x1b[0m')
if b == 'q':
print 'Bye Bye!!'
break
elif len(b) > 0:
raw = Parser.runtimeParser(b + ' ', SEQ_LEN)
question = []
question.append([wmap[c] for c in raw])
_input = np.array(question[0], np.int32)
_output = np.zeros([SEQ_LEN], np.int32)
tensorNet.inference(engine, _input, _output)
print 'Q: ' + '\x1b[1;39;94m' + Parser.decodeData(
question[0]) + '\x1b[0m'
print 'A: ' + '\x1b[1;39;92m' + Parser.decodeData(
_output.tolist()) + '\x1b[0m'
dims_h1 = engine.get_binding_dimensions(4).to_DimsCHW()
dims_c1 = engine.get_binding_dimensions(5).to_DimsCHW()
dims_ot = engine.get_binding_dimensions(10).to_DimsCHW()
output_ot = cuda.pagelocked_empty(dims_ot.C() * dims_ot.H() * dims_ot.W() * MAX_BATCHSIZE, dtype=np.float32)
d_in_enc = cuda.mem_alloc(MAX_BATCHSIZE * dims_enc.C() * dims_enc.H() * dims_enc.W() * _enc_text.dtype.itemsize)
d_in_dec = cuda.mem_alloc(MAX_BATCHSIZE * dims_dec.C() * dims_dec.H() * dims_dec.W() * _dec_text.dtype.itemsize)
d_in_h0 = cuda.mem_alloc(MAX_BATCHSIZE * dims_h0.C() * dims_h0.H() * dims_h0.W() * _h0.dtype.itemsize)
d_in_c0 = cuda.mem_alloc(MAX_BATCHSIZE * dims_c0.C() * dims_c0.H() * dims_c0.W() * _c0.dtype.itemsize)
d_in_h1 = cuda.mem_alloc(MAX_BATCHSIZE * dims_h1.C() * dims_h1.H() * dims_h1.W() * _h1.dtype.itemsize)
d_in_c1 = cuda.mem_alloc(MAX_BATCHSIZE * dims_c1.C() * dims_c1.H() * dims_c1.W() * _c1.dtype.itemsize)
d_ot_ot = cuda.mem_alloc(MAX_BATCHSIZE * dims_ot.C() * dims_ot.H() * dims_ot.W() * output_ot.dtype.itemsize)
bindings = [int(d_in_enc), int(d_in_dec), int(d_in_h0), int(d_in_c0), int(d_in_h1), int(d_in_c1),
int(d_in_h0), int(d_in_c0), int(d_in_h1), int(d_in_c1), int(d_ot_ot)]
stream = cuda.Stream()
while True:
b = raw_input('\n\n\x1b[1;105;97m'+'Please write your question(q for quite):'+'\x1b[0m')
if b=='q':
print 'Bye Bye!!'
break
elif len(b)>0:
raw = Parser.runtimeParser(b+' ', SEQ_LEN)
question = []
question.append([wmap[c] for c in raw])
question = np.array(question,dtype='int32')
print 'Q: '+'\x1b[1;39;94m'+Parser.decodeData(question[0])+'\x1b[0m'
print 'A: '+'\x1b[1;39;92m'+Parser.decodeData(inference(question[0]))+'\x1b[0m'