A hyper-fast local vector database for use with LLM Agents | Hacker News Reader

A hyper-fast local vector database for use with LLM Agents

github.com

74 pointslopkeny12ko21 comments

https://github.com/jdagdelen/hyperDB/blob/main/hyperdb/galax...

This was a great laugh. Also, after the benchmark, it says:

> Benchmark Credit: Benim Kıçım

"Benim Kıçım" means "my ass" in Turkish.

References "[Submitted on 31 Mar 2023] Programmed but Arbitrary Control Minimization of Amplitude and phase for speckle Nulling (PACMAN)" https://arxiv.org/abs/2303.18231

No idea if that's a joke paper.

> HyperSVMRanking (Such Vector, Much Ranking) algorithm proposed by Andrej Karpathy (2023) https://arxiv.org/abs/2303.18231

Damn this is good - well played.

For those who aren't neck deep in LLM hype - ChromaDB raised $18m at a $75m valuation a couple weeks ago with what's essentially just a thin wrapper around duckdb with a parquet import/export for persistence.

Hahaha. Really?! Good for them.

A "DB performance vs VC funding raised" plot is an odd choice for a github repo dealing with code.

I think that was perfectly on point if you reviewed the code.

Damn - did this get flagged? Went from top-of-HN to missing from the front page _pretty_ quick.

Yeah, this was a little strange. I'm the author of the repo linked here.

I got alerted to it hitting #1 on HN and was able to see it there for only a little bit before it immediately got tanked to # 60. I'm a little bummed, but c'est la vie.

Certainly suspicious - almost like VC's can't take a joke ;) I'm sure there's an innocent explanation though.

Great satire in any case! Strikes the perfect balance between highly plausible and glaringly obvious.

It’s a joke, but is it a functional one?

I found it crazy to run auto-GPT with an external DB over the internet, how can this be ever fast and efficient? (unless the dataset is humongous)

Yes, it used OpenAI embeddings and numpy to find cosine similarity

https://raw.githubusercontent.com/jdagdelen/hyperDB/main/_st...

Y-axis on this chart is actually hilarious

I couldn't believe they were raising $35 million rounds with a brand new embedding database that was open source. Then I poked through the dozens of lines of code and realized they had implemented the same DB I had.

LOL, much funny.

aside from making fun of Chroma, this seems to 100% work though? What would be the pros/cons of actually using this instead?

Author here. Yes, it actually does work. If I’m ever indexing fewer than ~100k documents/paragraphs I would use something like this approach rather than reaching for a vector database.

Great work!

`def derridaean_similarity`

Instant classic.

Absolutely hilarious. You're ready to raise $18 million.

It’s a prank.

Haha, so all one has to do is to add some fancy terms like LLMs, vectors, hyper-fast and 300+ Github stars. What a a joke lol!

Reply on news.ycombinator.com