Milvus Reviews

Milvus is primarily used in RAG, which involves retrieving relevant documents or data to augment the generation of new content. Milvus helps convert text and other data into a vector space, and the embeddings of this data are stored in the database. 

When a query is made, Milvus matches the query against the vector space to retrieve the most relevant vectors. For example, if we ask Milvus to retrieve the four best-matching vectors, it will provide those vectors. These vectors can then be decoded or reconverted into text, which can be further processed or used for generating new content.

Whenever we need to ask a language model a question, it can typically handle natural language queries effectively. However, if the model hasn't been trained on specific data related to the query, it may not be able to generate a precise answer.

This limitation is addressed by RAG, which allows the model to incorporate additional resources to answer questions beyond its training data. For organizations that use language models, if they have proprietary data such as PDFs, manuals, product roadmaps, or feature documents, RAG can help by providing this data to the model. This enables the model to answer questions related to that specific data by providing relevant context.

In our case, we've used Milvus to store organizational data, including PDFs and other data sources. Milvus facilitates the retrieval of relevant information to augment the model's ability to answer questions based on the stored data.

Milvus is an open-source vector database designed for efficiently handling large-scale, high-dimensional data. It supports various types of data sources and can be deployed on your own premises, which is crucial for maintaining data security.

Milvus offers multiple methods for calculating similarities or distances between vectors, such as L2 norm and cosine similarity. These methods help in comparing different vectors based on specific use cases. For instance, in our use case, we find that the L2 distance works best, but you can experiment with different methods to find the most suitable one for your needs.

Milvus also includes its own user interface, known as the Milvus Dashboard, which allows you to visualize and manage your data, including embeddings and metadata. You can filter your data based on various criteria, including metadata and file names, which provides flexibility in data management.

Milvus could be improved how it could automatically generate insights from the data it holds. Milvus maintains embedding information and knows the relationships between data points. It would be useful to explore how Milvus can provide more detailed insights or support additional business use cases, such as identifying clusters of similar data and deriving actionable insights from them. Additionally, video interpretation and protection could be areas where Milvus could offer extra value.

I have been using Milvus for a year.

Milvus provides documentation and an easy deployment process. Once you deploy Milvus, you can integrate it directly into your code. Using Milvus with Lambda is straightforward because it supports the necessary middleware version. You need to specify which database you're using, and Milvus will handle the connection automatically based on your configuration. 

Currently, our development team of over a hundred people manages eight to ten products running on Milvus.

If you are maintaining your own Milvus instance, you will need to provide the hardware and handle the management yourself. You’ll only need to pay for the resources you use, such as the machine and any associated costs. They do offer their own cloud instances, which should be manageable. You can explore their website to find detailed pricing information. For an on-premise instance, costs might be lower, but you’ll need to consider data growth and plan your budget accordingly. 

Milvus works well for various use cases and is quite flexible in terms of deployment. For on-premises deployment, you can use the open-source version with Docker. The system requirements are relatively modest; around 16 GB of RAM and some disk space are recommended. This setup is sufficient for initial trials or proof-of-concept projects.

If you prefer, you can use a cloud instance for Milvus. Cloud instances have limitations but are ideal for initial testing and quick setup without needing to install everything locally. This allows you to experiment with Milvus without significant upfront investment in infrastructure.

Milvus provides options for both on-premises and cloud deployment, so you can choose based on your needs. The documentation is comprehensive, and while some initial setup may require assistance, the process is straightforward once you get the hang of it. You can configure Milvus with different databases and customize it to fit your requirements.

Milvus excels at calculating the distance between your queries and the data, which is central to its functionality. The tool is designed to be easy to use and configure, with options to filter and view your data effectively. If you encounter issues, the UI and documentation provide support to help you troubleshoot and resolve problems.

Overall, I rate the solution a seven out of ten.

We use Milvus mostly for RAG (Retrieval-augmented generation).

Since Milvus is an open-source solution, I can use it without worrying about licenses or costs. The solution is well containerized, and since containerization is quick and easy for me, I can scale it up quickly.

Milvus' documentation is not very user-friendly and doesn't help me get started quickly. Milvus usually comes with some dependencies because of the way it needs to be deployed.

I have been using Milvus for around one year. 

Milvus is a stable solution.

Around 20 users are using the solution in our organization.

Milvus' documentation is not very user-friendly and doesn't help me get started quickly. Chroma DB provides super user-friendly documentation, enabling new users to get started quickly. Chroma DB's setup doesn't have many dependencies, whereas Milvus usually comes with some dependencies because of the way it needs to be deployed. Unlike Milvus, it's very easy to do POCs with Chroma DB.

Milvus is powerful and can be used by an experienced developer or architect. A junior developer or somebody new to Python, AI applications, containers, or dockers will take time to set up and use the solution.

Milvus is an open-source solution. I haven't encountered any hidden fees for the solution, but I believe they provide some paid support.

I work in a services company that works with different customers. Often, customers decide to use Milvus rather than us. I would recommend Milvus to experienced software developers.

Local infrastructure or on-premises will use Mistral AI for Large Language Models (LLMs)​ and Chroma DB for vector DB. We usually use the BGE, LangChain, or LlamaIndex embedding models.

Overall, I rate the solution a six out of ten.

Initially, I used it with software support and related products. Personally, I installed it locally on Docker containers for testing.

I used it for data storage and search queries, mainly for sharing data across different selections. I used Milvus to create datasets, each with different scenarios or purposes.

I like the accuracy and repository. They provide everything very precisely, including all of their installation processes. Their service providers and parameters, especially the dimension parameters, make the software incredibly user-friendly for both beginners and experienced users. This is especially helpful since vector stores can be tricky. Milvus' system allows for continuous learning and improvement, which we can then pass on to our customers. 

I also appreciate the availability of filter conditions, search speed, and all the features that help us learn and use it effectively.

It's easy to learn. Once you understand the different features, Dokcer and data analytics capabilities, you can use it effectively for your resources.

There are some requirements, but if users understand how to load data and other basic functions, they can integrate it easily.

It doesn't take too much time to learn if someone is motivated to learn. However, if they are unfamiliar with programming and related concepts, it could take two to six months.

Milvus could make it simpler. Simplifying the requirements and making it more accessible. It could be more user-friendly. 

However, their recent addition of a support server and their ongoing work indicate a commitment to continuous improvement.

I used it in the past year.

It is a stable product. It does not break down. There are no issues. 

I primarily used it myself, but we conducted extensive research as a team.

We explored different options like local and personal servers, but Milvus's resources and impact made it the best solution. I also tried other vector databases, but their focus on local users didn't align with our needs.

Installing Milvus can be tricky because we have to install Docker. For deployment, ensuring compatibility requires using different platforms. It might slow down the process. The installation process and everything provided in the documentation is helpful, but it can still be a bit complex.

Not too complex if done properly, but new users may find it difficult.

It's an open-source tool. We used it locally for research, so we didn't have to pay anything. 

I would definitely recommend it. For all the things it provides, it is a good solution. 

Overall, I would rate the solution a nine out of ten. 

We were building an FAQ solution. When we had too many documents, we created chunks of those documents and stored them in the Milvus database. When someone asked any questions present in the document, we would create an embedding of that document, and we found them in our chunks stored in Milvus. This way, we got the top three results or chunks from Milvus.

The best feature of Milvus was finding the closest chunk from a huge amount of data.

I've heard that when we store too much data in Milvus, it becomes slow and does not work properly.

I have been using Milvus for around one year.

Around eight to ten teams were working with the solution in our organization.

Milvus is an open-source solution.

We were deploying Milvus using Docker on our AWS. Recommending Milvus to other users depends on their data size and chunk amounts. Based on those parameters, users can decide whether to go with Milvus or find other options, like OpenSearch from AWS. New users can easily learn to use Milvus for the first time by following the documentation page. It was easy to integrate Milvus with our existing services like Python.

Overall, I rate the solution an eight out of ten.

I use Milvus mostly for text processing and natural language processing.

Milvus has good accuracy and performance.

Milvus has more overhead than other solutions especially with etcd setup and is too heavy for our use cases. Milvus has higher resource consumption, which introduces complexity in implementation.

I have been using Milvus for 4 years.

I rate Milvus a six out of ten for scalability.

I previously used Elasticsearch. I switched to Milvus because I had reached the end of capabilities  Elastic Search as Indexes were getting quite large. Certain use cases would have too much data to search. We switched to Milvus because our use cases were becoming too slow so we opted for Vector Space instead.

The solution's initial setup could be complex for large-scale deployment.

The solution's deployment takes two to three hours, plus testing and monitoring time.

We have seen a return on investment with Milvus.

I tried Weaviate at the time and opted for Milvus because it was more straighforward for our use case.

Milvus is deployed on-cloud in our organization.

Overall, I rate Milvus a seven out of ten.