groot
groot
你用的是python sdk，我说过python语言对于浮点型的数值总是转成64位显示。 如果你懂得怎么在python sdk里调试的话，你会看到query()接口会调用到这里：https://github.com/milvus-io/pymilvus/blob/473f62fee6cb5e446f571471884d0d7303d01ab6/pymilvus/client/entity_helper.py#L481 我们把这几行代码改成如下： ``` if field_data.type == DataType.FLOAT_VECTOR: dim = field_data.vectors.dim if len(field_data.vectors.float_vector.data) >= index * dim: start_pos, end_pos = index * dim, (index + 1) * dim...
**再强调一遍，milvus服务端处理向量，无论是存储，计算还是输出，都是以float32来处理。各个SDK拿到的向量数据都是32位浮点型。如果你看到的是64位浮点数，那是因为某些语言本身自动把32位转成了64位表示，实际上32位精度之外的数字都是随机数字，没有任何意义。**
Milvus v2.4.15中已支持数组索引： https://milvus.io/docs/array_data_type.md#Index-array-field
I have no idea. Might related to the local Java environment.
What do you mean "io.milvus.grpc does not exist"? 
已经有人把milvus java sdk用在springboot里了。MilvusServiceClient底下依赖于grpc的实现，所有远程接口的调用由grpc保证其线程安全性，可以用多线程并发。
Maven is recommended, we didn't provide a build script for Gradle. For the error "package io.milvus.grpc does not exist", the answer is in this issue: https://github.com/milvus-io/milvus-sdk-java/issues/273
The tests launch a milvus server locally. You can see there is a docker-compose.yml under the java sdk project root path. The master&v2.3 branch uses milvus v2.3.0 to test, the...
I guess your milvus is not 2.4.14. Both the pymilvus 2.4.9 and java sdk 2.4.7 work very well with milvus 2.4.14.
Show me the test.py