redskyluan

Maybe check this https://zilliz.com/pricing

Yan easily store 1B data into Zilliz Serverless and the cost is incredible cheap

Interesting case study, but I'm skeptical of the broader implications. Pinecone is notoriously expensive compared to other vector database services. A horizontal price comparison reveals several better options in the market. Removing the calculator doesn't solve the core issue - it just obfuscates costs and makes it harder for users to compare options upfront. In my view, this approach merely reduces the comparison stage, potentially leading more uninformed users to upload data without fully understanding the pricing implications. While simplification can be valuable, in this case it seems to benefit the company more than the users. A better approach might be to improve the calculator's accuracy and usability rather than removing it entirely. Transparency in pricing is crucial, especially for B2B services where costs can scale quickly.

there seems to be a big hype to adapt pg into any infra. I love PG but this seems not be right thing.

Based on your use cases. I thought it's not hard to push the window to 32K or even 100k if we change the position embedding

Something need to be fixed. Opensource is not about license, it's about people.

OSS companies are fighting against cloud vendors to survice.

https://zilliz.com/blog/Redis-tightens-its-license-How-can-a...

What I Observe: Simple RAG is Fading, but Complex RAG Will Persist and Evolve - Involving Query Rewriting, Data Cleaning, Reflection, Vector Search, Graph Search, Rerankers, and More Intelligent Chunking. Large Models Should Not Just Be Knowledge Providers, But Tool Users and Process Drivers"

I would personally not recommend implementing a Redis protocol on top of SQLite, as I've seen too many failed cases like this. Users may perceive your product as a drop-in Redis replacement, and it may work fine in a PoC, but once it hits production, SQLite's performance and scalability will be severely challenged.

It's much better to use RocksDB as the underlying storage engine for such a solution. RocksDB can provide the performance and scalability required.If you need a distributed solution, I would suggest looking at TiKV or FoundationDB. These are both excellent distributed storage systems that can handle production workloads much better than a SQLite-based approach.

finally we see some relationship between Golang and AI?

I'm tired about the hype of python and rust. I would really appreciate if we saw more AI project written in golang.

for example: https://github.com/tmc/langchaingo

What an article!

GResNet seems to generated the most similar result from my human evaluation.

Maybe we can combine both model with a hybrid search?

what will be the next step of 1 bit? could it be 0.5 bit? Any kind of quantization is not gonna to last long.

By the way, we are looking for talents of velox, data fusion, ducdb, clickhouse to build the world's fastest vector db milvus(mainly on search side). Contact me if you have this background!

Velox could be competitor of datafusion. It is more focus on execution engine and could be great to integrate to other high performance databases.

Database will be split into pieces and rebuild!

Great product and great progress.

The first step to build rage is always to evaluate.

Except all the current evaluations, cost and perf should also be part of evaluations

This seems not be a repo ready to open source. You only get weights, very less information about how the weights is trained and finetuned.

But anyway, it always great to see more LLM weigts available.

Lessons learned:

The Importance of Model Agnosticism: With the rapid evolution of AI models, building applications that are model-agnostic has become more critical than ever.

Control and Interpretability Matter: Relying solely on large language models (LLMs) poses significant challenges for creating applications that can be deployed in real-world scenarios.

The Need for Open Models: Lastly, the push for more open models has never been more apparent. Open-source models are essential for fostering innovation, ensuring accessibility, and maintaining the integrity of our work in the AI field.

As an open source database Milvus contributor, I am deeply grateful for the contributions brought by all non-code contributors to the project. However, in my view, this is far from sufficient. Without the project maintainers or dedicated documentation developers to operate the official website, you will soon find the documentation filled with a lot of outdated information, misconceptions, and some incorrect understandings. More crucially, if core contributors do not deeply participate and draft the first version of the functional documentation, it is likely to be very difficult to read because it would lack the mental model documents that explain how the authors think about things. In fact, aside from simple user documentation, a good FAQ can help others better understand the design, and this must be the responsibility of the core developers.

Regarding scale, PostgreSQL may not cover all bases. Though I'm a PostgreSQL fan, I prefer specialized services for specific tasks. Using PG-based plugins could help, but a dedicated SQL-compatible database is often a better fit.

For vector retrieval, going with a database like Milvus, designed for vectors, is usually more efficient and cost-effective. Similar principles apply across domains. Is there any vectordb under PG format?

What if we've got a deeply customized distributed vector search service on PostgreSQL, that's impressive!

this is much cheaper than S3 glacier. Any explanation on how to keep data safe?

We also built a similar semantic search engine for retrieving open-source projects. We experimented with HyDE and guessing user queries based on a corpus, both of which were very successful.

For embeddings, we chose BGE, but found they seemed to overfit on Beir. CohereV3 and Voyage appeared to perform better in practice.

For retrieval, we used OpenSearch + ZillizCloud Serverless + Cohere ranking, providing us with maximum flexibility and search effectiveness.

To be mentioned, build a evaluation dataset is important. This helps me to improve the quality.

Just stumbled upon this blog it's absolutely intriguing! As a Python enthusiast, it's like finding a hidden treasure that challenges the usual norms of Python's capabilities. Thinks of writing a Pure python implementation of some ml algos in learning SIMD~

The inclusion of the OpenAI dataset in this benchmark adds a layer of realism that's often missing in standardized tests with datasets like SIFT and DEEP

I think we need vectordb bench on 100M level. If you don't have 100M data, and you don't care about things like filtering and streaming insertion, I vote for PGVector since SQL is convenient enough.

However, for large dataset deployment, cost becomes more critical since vector search is computation intensive. Anything like es, mongodb and redis can not even share their results in the benchmark.

Also, if you are looking for more fancy features rather simply ANN, purpose built vector database has faster iterations than traditional databases

I think user has to test by themselves. Vectorbenchmark support you to run the test by yourself on any cloud serivce or opensource deployment. One of my guess is qdrant tune their parameters crazily on their benchmark.

how do you define chroma as a vector db? a wrapper on top of other databases and hnsw? What about a billion vectors in house~ is chroma all we need?

That's exactly what I'm trying to do. Play with prompts, generate multiple questions and cluster questions and pick some of the centroid questions to do embeddings

They generate questions like:

where did you go this morning? When did you woke up this morning. What did you do after breakfast? What did you do today at Golden Gate Park.

GPT is all about probabilities. So the LLM know what might be most related answer of a doc chunk.

It works much better than embedding the whole sentence because "When did you woke up this morning" might not be very similar with "Today I woke up at 9.am, had a light breakfast and then went on a run in Golden Gate Park.".

i have an opposite way on doing this. Tried to generate questions based on doc chunks and embedding on questions. It works perfect!

I would vote for finetuning, prompt engineering, rather than only add domain specific knowledge. Others are playing detective ,digging into the ethical conundrums of AI-generated content.

Can't Agree with that more.

LLMs should not be trained to simply memorize information. Instead, they should be designed to understand and identify patterns in the data, and use the knowledge stored in vector databases to organize and summarize information.

Vector databases can be used to store and organize knowledge in a way that is more accessible to LLMs. By using vector representations, LLMs can easily access and manipulate knowledge, allowing them to more effectively process and analyze large amounts of information.

OSSChat is a free chatbot that helps you learn more about open source software. It is powered by ChatGPT, Langchain, and Milvu and supported community like pytorch, huggingface and ray.

We are pleased to announce that we now support OpenCV. Feel free to ask any questions about the project, and OSSChat will help you find relevant documents to assist you. Enjoy!

We welcome your comments on how we can improve OSSChat. Nominations of new open source community are also welcomed.