ERINE-doc调用

[xiankangW-gi]

在使用demo时：
import numpy as np
import paddle
from paddlenlp.transformers import ErnieDocModel
from paddlenlp.transformers import ErnieDocTokenizer

def get_related_pos(insts, seq_len, memory_len=128):
beg = seq_len + seq_len + memory_len
r_position = [list(range(beg - 1, seq_len - 1, -1)) +
list(range(0, seq_len)) for i in range(len(insts))]
return np.array(r_position).astype('int64').reshape([len(insts), beg, 1])

tokenizer = ErnieDocTokenizer.from_pretrained('ernie-doc-base-zh')
model = ErnieDocModel.from_pretrained('ernie-doc-base-zh')

inputs = tokenizer("长文本")
inputs = {k:paddle.to_tensor([v + [0] * (128-len(v))]).unsqueeze(-1) for (k, v) in inputs.items()}

memories = [paddle.zeros([1, 128, 768], dtype="float32") for _ in range(12)]
position_ids = paddle.to_tensor(get_related_pos(inputs['input_ids'], 128, 128))
attn_mask = paddle.ones([1, 128, 1])

inputs['memories'] = memories
inputs['position_ids'] = position_ids
inputs['attn_mask'] = attn_mask

outputs = model(**inputs)

encoder_output = outputs[0]
pooled_output = outputs[1]
new_mem = outputs[2]
请问ERINE-doc不是号称可以处理长文本吗 为什么在当输出文本超过512tokens时 依然出现：Token indices sequence length is longer than the specified maximum sequence length for this model (566 > 512). Running this sequence through the model will result in indexing errors