def run(self, record):
text_a, text_b = record[0], record[1]
example = self.processor._create_single_example(text_a, text_b)
feature = convert_single_example(example, 32, self.tokenizer)
input_ids = torch.tensor(feature.input_ids,
dtype=torch.long).unsqueeze(0)
segment_ids = torch.tensor(feature.segment_ids,
dtype=torch.long).unsqueeze(0)
input_mask = torch.tensor(feature.input_mask,
dtype=torch.long).unsqueeze(0)
# ort_inputs = {
# 'input_ids': self.to_numpy(input_ids),
# 'segment_ids': self.to_numpy(segment_ids),
# 'input_mask': self.to_numpy(input_mask)
# }
# print(input_ids)
# print(segment_ids)
# print(input_mask)
ort_inputs = {
'input_ids': self.to_numpy(input_ids),
'segment_ids': self.to_numpy(segment_ids),
'input_mask': self.to_numpy(input_mask)
}
print(self.session)
ort_outputs = self.session.run(None, ort_inputs)
# print(ort_outputs)
ort_logits = torch.from_numpy(ort_outputs[0])
print(ort_logits) # tensor([[4.7433, -4.5335]])
# ort_logits_1 = torch.from_numpy(ort_outputs_1[0]) # 二维向量
prob = ort_logits.sigmoid()[:, 1].tolist()[0] #[0.123]
# print(ort_logits)
return prob
def run(self, record):
'''
预测小类标签
'''
text_a, text_b = record[0], record[1]
example = self.processor._create_single_example(text_a, text_b)
feature = convert_single_example(example, self.max_seq_length,
self.tokenizer)
input_ids = torch.tensor(feature.input_ids,
dtype=torch.long).unsqueeze(0).to(
self.config.device)
segment_ids = torch.tensor(feature.segment_ids,
dtype=torch.long).unsqueeze(0).to(
self.config.device)
input_mask = torch.tensor(feature.input_mask,
dtype=torch.long).unsqueeze(0).to(
self.config.device)
# print(input_ids)
# print(segment_ids)
# print(input_mask)
logits = self.model(input_ids, segment_ids, input_mask).detach()
# print(logits)
prob = logits.sigmoid()[:, 1].tolist() #[0.123]
# prob = torch.sigmoid(logits)
# return prob[0].cpu().tolist()[0]
return prob[0]
def infer(bert, nezha, query_A, query_B):
text_a, text_b = query_A, query_B
example = processor._create_single_example(text_a, text_b)
feature = convert_single_example(example, max_seq_len, tokenizer)
input_ids = torch.tensor(feature.input_ids, dtype=torch.long).unsqueeze(0)
segment_ids = torch.tensor(feature.segment_ids,
dtype=torch.long).unsqueeze(0)
input_mask = torch.tensor(feature.input_mask,
dtype=torch.long).unsqueeze(0)
ort_inputs = {
'input_ids': to_numpy(input_ids),
'segment_ids': to_numpy(segment_ids),
'input_mask': to_numpy(input_mask)
}
ort_outputs_bert = bert.run(None, ort_inputs)
ort_outputs_nezha = nezha.run(None, ort_inputs)
ort_logits_bert = torch.from_numpy(ort_outputs_bert[0])
ort_logits_nezha = torch.from_numpy(ort_outputs_nezha[0])
res = torch.mean(torch.stack([ort_logits_bert, ort_logits_nezha]), 0)
prob = res.sigmoid()[:, 1].tolist()[0]
return prob
def infer(model, query_A, query_B):
text_a, text_b = query_A, query_B
example = processor._create_single_example(text_a, text_b)
feature = convert_single_example(example, max_seq_len, tokenizer)
input_ids = torch.tensor(feature.input_ids,
dtype=torch.long).unsqueeze(0).to(config.device)
input_mask = torch.tensor(feature.input_mask,
dtype=torch.long).unsqueeze(0).to(config.device)
segment_ids = torch.tensor(feature.segment_ids,
dtype=torch.long).unsqueeze(0).to(config.device)
logits = model(input_ids, segment_ids, input_mask).detach()
prob = logits.sigmoid()[:, 1].tolist() #[0.123]
return prob[0]
def infer(model, query_A, query_B):
text_a, text_b = query_A, query_B
example = processor._create_single_example(text_a, text_b)
feature = convert_single_example(example, max_seq_len, tokenizer)
input_ids = torch.tensor(feature.input_ids, dtype=torch.long).unsqueeze(0)
segment_ids = torch.tensor(feature.segment_ids, dtype=torch.long).unsqueeze(0)
input_mask = torch.tensor(feature.input_mask, dtype=torch.long).unsqueeze(0)
ort_inputs = {
'input_ids': to_numpy(input_ids),
'segment_ids': to_numpy(segment_ids),
'input_mask': to_numpy(input_mask)
}
ort_outputs = model.run(None, ort_inputs)
ort_logits = torch.from_numpy(ort_outputs[0])
prob = ort_logits.sigmoid()[:, 1].tolist()[0] #[0.123]
return prob
def run(self, record):
text_a, text_b = record[0], record[1]
example = self.processor._create_single_example(text_a, text_b)
feature = convert_single_example(example, self.max_seq_length,
self.tokenizer)
input_ids = torch.tensor(feature.input_ids,
dtype=torch.long).unsqueeze(0)
input_mask = torch.tensor(feature.input_mask,
dtype=torch.long).unsqueeze(0)
segment_ids = torch.tensor(feature.segment_ids,
dtype=torch.long).unsqueeze(0)
ort_inputs = {
'input_ids': input_ids,
'input_mask': input_mask,
'segment_ids': segment_ids
}
ort_outputs = self.session.run(None, ort_inputs)
print(ort_outputs)
print(type(ort_outputs))
