How does it work if we have both Cursive and clojure-lsp running? Should we stop using clojure-extras if we're using this?

This is so good news.

@UEENNMX0T unfortunately, that doesn't work with cursive, more context https://clojurians.slack.com/archives/C04CAKAGADU/p1690147553580799, but I'd say the static analysis features it provides should be better than clojure extras as will use clojure-lsp which has all kondo integration and more like unused-public-var linter and clj-depend integration.

Incredible timing, yesterday I was researching java bindings for llama.cpp :-)

Come join us in #C054XC5JVDZ!

Thanks for putting the llama.clj library together and congrats on the release @U7RJTCH6J! I've been using your early versions of llama.clj for a few weeks and its been fantastic. Here are a few of my observations so far: The library has very low overhead - so the performance is on par with llama.cpp - which is one of the faster LLM libraries out there. There was no python to setup which simplified the setup significantly. Llama.clj works with Nvidia GPUs and Apple Silicon etc - but also works to a certain extent with just an x86 CPU too (around 5 tokens/second on an oldish AMD CPU) which can work for certain usecases. Having an LLM available as a library in Clojure really changes how you tackle programming problems since you can now delegate tasks to the LLM with no dependency on third party APIs (so no credit card needed or managing secret API credentials etc). Look forward to expanding my usage of it over time!

I must be doing something wrong: • I've added the precompiled lib to deps.edn • started the repl, and (require '[com.phronemophobic.llama :as llama]) and I'm getting a class not found error

There’s at least two libs to add to your deps Edn I think.

oh, let me have a look

duh, I was missing the clj-lama lib

Let me know if there’s an edit to the Readme that might make things more clear!

Yes, I think I just assumed that using precompiled native dependencies will pull in the main lib

native libs are kind of a pain and I still haven’t figure out a way to make them easy to include.

I that's kinda expected, given the nature of the library - when I worked with CoreNLP it worked in the same way, although there was some Maven magic they pulled off to make default installation more usable

@U7RJTCH6J Awwwright, got it working in the REPL - I can open a PR with small tweaks to the readme if you want

Yea, that would be great!

Done!

Bravo!