def forward(self, q, a_batch):
# bs, seq, Hidden*2
q = self.lstm_q(q)
# bs, 1, seq, hid*2
q = autograd.reshape(q, (q.shape[0], 1, q.shape[1], q.shape[2]))
# bs, 1, 1, hid*2
q = self.q_pool(q)
# bs, hid*2
q = autograd.reshape(q, (q.shape[0], q.shape[3]))
# 2bs, seq, Hidden*2
a_batch = self.lstm_a(a_batch)
# 2bs, 1, seq, hid*2
a_batch = autograd.reshape(
a_batch, (a_batch.shape[0], 1, a_batch.shape[1], a_batch.shape[2]))
# 2bs, 1, 1, hid*2
a_batch = self.a_pool(a_batch)
# 2bs, hid*2
a_batch = autograd.reshape(a_batch,
(a_batch.shape[0], a_batch.shape[3]))
# 2*(bs, hid*2)
a_pos, a_neg = autograd.split(a_batch, 0, [q.shape[0], q.shape[0]])
sim_pos = autograd.cossim(q, a_pos)
sim_neg = autograd.cossim(q, a_neg)
return sim_pos, sim_neg
def forward(self, q, a_batch):
q = autograd.reshape(q, (q.shape[0], -1)) # bs, seq_q*data_s
a_batch = autograd.reshape(a_batch,
(a_batch.shape[0], -1)) # 2bs, seq_a*data_s
q = self.linear_q(q) # bs, hid_s
a_batch = self.linear_a(a_batch) # 2bs, hid_s
a_pos, a_neg = autograd.split(a_batch, 0,
[q.shape[0], q.shape[0]]) # 2*(bs, hid)
sim_pos = autograd.cossim(q, a_pos)
sim_neg = autograd.cossim(q, a_neg)
return sim_pos, sim_neg
def forward(self, x):
y = self.lstm(x)
y = autograd.reshape(y, (y.shape[0], -1))
y = self.l1(y)
y = autograd.relu(y)
y = self.l2(y)
return y
def predict(self, queries: List[str]):
print("Get queries")
res = []
queries = [self._preprocess(ele) for ele in queries]
for input_ids in queries:
x = tensor.Tensor(device=self.dev, data=input_ids)
out = []
for i in range(self.length):
y = self._model.forward(x)
y = autograd.reshape(y, y.shape[-2:])[-1, :]
y = tensor.softmax(y)
y = tensor.to_numpy(y)[0]
y = np.argsort(y)[-1]
out.append(y)
y = np.array([y]).reshape([1, 1, -1]).astype(np.float32)
y = tensor.Tensor(device=self.dev, data=y)
x = tensor.concatenate([x, y], 2)
result = self._postprocess(out)
res.append(result)
return res
def forward(self, x):
y = self.conv1(x)
y = self.bn1(y)
y = autograd.reshape(y, (y.shape[0], -1))
y = self.doublelinear1(y)
return y
def forward(self, x):
y = self.lstm(x)
if self.return_sequences:
y = autograd.reshape(y, (-1, self.seq_length * self.hidden_size))
return y
def forward(self, x, seq_lengths):
y = self.lstm(x, seq_lengths=seq_lengths)
y = autograd.reshape(y, (y.shape[0], -1))
y = self.l1(y)
return y
logging.info("model compling...")
dev = device.get_default_device()
x = tensor.Tensor(device=dev, data=input_ids)
model = MyModel(onnx_model)
# verifty the test
# from utils import load_dataset
# sg_ir = sonnx.prepare(onnx_model) # run without graph
# inputs, ref_outputs = load_dataset(
# os.path.join('/tmp', 'GPT-2-LM-HEAD', 'test_data_set_0'))
# outputs = sg_ir.run(inputs)
# for ref_o, o in zip(ref_outputs, outputs):
# np.testing.assert_almost_equal(ref_o, o, 4)
logging.info("model running...")
output = []
for i in range(length):
logging.info("word {} generating...".format(i))
y = model.forward(x)
y = autograd.reshape(y, y.shape[-2:])[-1, :]
y = tensor.softmax(y)
y = tensor.to_numpy(y)[0]
y = np.argsort(y)[-1]
output.append(y)
y = np.array([y]).reshape([1, 1, -1]).astype(np.float32)
y = tensor.Tensor(device=dev, data=y)
x = tensor.concatenate([x, y], 2)
text = postprocess(output)
print(text)
def loss(self, out, ty):
ty = autograd.reshape(ty, (-1, 1))
return autograd.softmax_cross_entropy(out, ty)
def forward(self, inputs):
x, self.hx, self.cx = self.rnn(inputs, (self.hx, self.cx))
x = autograd.cat(x)
x = autograd.reshape(x, (-1, self.hidden_size))
return self.dense(x)