#Query #transformers model #neuralnetwork weighs as a #graph #database. And surgically insert new information into the model. The #model is the #DB!
#OpenSource Apache 2.0 lic
#AI #LLM

Gotta say: pretty cool ! You can decompose models into a graph database 👍️👍️
old.reddit.com/r/MachineLea...

GitHub - chrishayuk/larql #larql
github.com/chrishayuk/l...

LLMs Are Databases - So Query Them
www.youtube.com/watch?v=8Ppw...

#LLM #GraphDatabases #SQL

@rhizonymph.com chris hay (chrishayuk on github/x) — read and write features, compile edits back to weights. natural counterpart to your steering runtime, and since you do open-source interp figured you'd want it on your radar.
video: www.youtube.com/watch?v=8Ppw...
repo: github.com/chrishayuk/larql

@rhizonymph.com saw your vLLM steering work — the magnitude-control problem really rhymes with LARQL: it decomposes the Gemma FFN into a queryable gate×down graph and does training-free fact INSERT via a balancer-scaled triple. same problem from the authoring side 🧵

https://github.com/chrishayuk/larql

Unveiling LARQL: Transforming Transformers Into Queryable Graph Databases Imagine every large language model you’ve ever interacted with—now consider for a moment that each isn’t simply a mer...

#Large #Language #Models #AI #interpretability #chris #hay #chrishayuk #feature #visualization #feed […]

Updated parameters of the LArQL model
http://arxiv.org/abs/2504.17866
L. Paulucci, F. Cavanna, V. Vale, F. Marinho

Came across LARQL earlier - imagine you could query the language and knowledge inside an LMM like its a Database, because it is actually just a database.
Video prompt: youtu.be/8Ppw8254nLI?...
Ive not had a chance to explore this yet ....

github.com/chrishayuk/l...

LARQL - Query neural network weights like a graph database
Discussion | lobsters | Author: sylq

Perfect database doesn't exi...

https://github.com/chrishayuk/larql

detailed mathematical account of the process is still to be achieved, a phenomenological model for simultaneously treating ionization and scintillation, LArQL, has been successfully employed to describe the range of electric fields from 0 to 0.75 [2/3 of https://arxiv.org/abs/2504.17866v1]

L. Paulucci, F. Cavanna, V. Vale, F. Marinho: Updated parameters of the LArQL model https://arxiv.org/abs/2504.17866 https://arxiv.org/pdf/2504.17866 https://arxiv.org/html/2504.17866

LARQL - Query neural network weights like a graph database https://lobste.rs/s/iawjcg ##databases ##ai

Just thinking about the fact that LLMs are 'just' databases, and that clients like LARQL can be used to query them with basic SQL.

So, why not use LARQL to extract/'scrape' useful/valuable data back out of the models? (To then use in regular apps etc.)

Hmm....

New on InsiderLLM: LARQL argues the FFN inside your transformer is already a graph database.

Extract it, patch it with 10KB files, walk it with KNN instead of dense matmul. Run bigger models on modest hardware.

Compelling idea. Author-only benchmarks. Honest take:

insiderllm.com/guides/run-3...