执行推断时出错TypeError: For primitive[Dense], the input type must be same.

[Dylan-JD]

If this is your first time, please read our contributor guidelines: https://github.com/mindspore-lab/mindcv/blob/main/CONTRIBUTING.md

Describe the bug/ 问题描述 (Mandatory / 必填) 执行推断时报错： TypeError: For primitive[Dense], the input type must be same. name:[w]:Ref[Tensor[Float32]]. name:[x]:Tensor[Float16].

• Hardware Environment(Ascend/GPU/CPU) / 硬件环境: Ascend 910B3
• Software Environment / 软件环境 (Mandatory / 必填): Mindspore 20231215
• Excute Mode / 执行模式 (Mandatory / 必填)(PyNative/Graph): Graph To Reproduce / 重现步骤 (Mandatory / 必填) Steps to reproduce the behavior: 使用bert类型的model，先用tokenizer进行encode，随后使用Tensor转为Tensor，最后使用model(Tensor)获得logistics

Expected behavior / 预期结果 (Mandatory / 必填) 输出一个embedding序列 Screenshots/ 日志 / 截图 (Mandatory / 必填)

Additional context / 备注 (Optional / 选填) Add any other context about the problem here.

[lvyufeng]

from mindnlp.models import MSBertModel
import mindspore
from mindnlp.transformers import BertTokenizer
from mindspore import Tensor
import json
import numpy as np
from gaussdb_vector.gauss_db_vector import GaussDBVector
from gaussdb_vector.gauss_table import FieldSchema, IndexSchema, TableSchema
from gaussdb_vector.gauss_types import DataType, MetricType, IndexType, IndexParamType, IndexParams
import sys
sys.path.append('/home/workspace/dongjian/gauss_test/code/gauss_python_program/gaussdb_vector')

# tokenizer = BertTokenizer.from_pretrained('/home/ma-user/data/bert_base_chinese')
tokenizer = BertTokenizer.from_pretrained('/home/workspace/dongjian/gauss_test/code/m3e_model')
mindspore.set_context(mode=mindspore.GRAPH_MODE)
# model = MSBertModel.from_pretrained('/home/ma-user/data/bert_base_chinese')
model = MSBertModel.from_pretrained('/home/workspace/dongjian/gauss_test/code/m3e_model')
model = model.to_float(mindspore.float16)

def predict(text, label=None):
    # print(tokenizer.encode(text))
    # if len(tokenizer.encode(text).idx) >= 512:
    if len(tokenizer.encode(text)) >= 512:
        return None
    # text_tokenized = Tensor([tokenizer.encode(text).ids])
    print(tokenizer.encode(text))
    text_tokenized = Tensor(np.array(tokenizer.encode(text)))
    print(type(text_tokenized))
    logits = model(text_tokenized)
    print(logits[0].shape)
    embedding = np.mean(logits[0][0].asnumpy(), axis=0)
    embedding = embedding.tolist()

    return embedding

print(predict('你好'))