Llama-2-Onnx VS libsql

System: Here's some docs, answer concisely in a sentence.

YMMV on cost still, depends on cloud vendor, and my intuition & viewpoint agrees with yours, GPT-3.5 is priced low enough that there isn't a case where it makes sense to use another model.

It strikes me now that _very_ likely and not just our intuition: OpenAI's $/GPU hour is likely <= any other vendor's.

The next big step will come from formalizing the stuff rolling around the local LLM community, for months now it's either been one-off $X.c stunts that run on desktop, and the vast majority of the _actual_ usage and progress is coming from porn-y stuff, like all nascent tech.

Microsoft has LLaMa-2 ONNX available on GitHub[1]. There's budding but very small projects in different languages to wrap ONNX. Once there's a genuine cross-platform[2] ONNX wrapper that makes running LLaMa-2 easy, there will be a step change. It'll be "free"[3] to run your fine-tuned model that does as well as GPT-4 .

It's not clear to me exactly when this will occur. It's "difficult" now, but only because the _actual usage_ in the local LLM community doesn't have a reason to invest in ONNX, and it's extremely intimidating to figure out how exactly to get LLaMa-2 running in ONNX. Microsoft kinda threw it up on GitHub and moved on, the sample code even still needs a PyTorch model. I see at least one very small company on HuggingFace that _may_ have figured out full ONNX.

[1] https://github.com/microsoft/Llama-2-Onnx

> PS: I've got nothing against Turso, or libSQL. In fact I spent the last year perusing their virtual WAL API. The problem is that I found no documentation, nor any useful open source implementations of it. If there any I'd be very interested. So, thus far, I also don't have anything that drives towards libSQL.

Hey, this is v and I am an engineer at Turso. We do have some documentation and an example implementation of Virtual WAL

docs: https://github.com/tursodatabase/libsql/blob/ef44612/libsql-...

example: https://github.com/tursodatabase/libsql/blob/ef44612/libsql-...

for an open source implementation, you may check how Bottomless works. Bottomless is another project which does back up like litestream and it internally implements a Virtual WAL.

Bottomless - https://github.com/tursodatabase/libsql/tree/main/bottomless

I am sure we can improve our docs, make it more discover-able and easy to find. I am open to feedback and suggestions!

Astro DB is powered by LibSQL, an open source fork of SQLite that was created by Turso. You can use Astro DB's drop-in database to build features like blogs, comment functionality, forums, feedback systems, and user authentication.

sqlite already had an active community fork started by Turso called libsql. They are fixing longstanding API gaps the upstream team isn’t interested in supporting. For example, they added a native write-ahead log API, so you can plug directly into the WAL for streaming replication. This is possible-ish with upstream sqlite + LiteFs but litefs has to implement a whole FUSE file system and can’t run on Mac for that reason.

It’s more risky to run libsql because new features mean new bugs, but it seems worth it to me.

Libsql: https://github.com/tursodatabase/libsql

A bit of a tangent but for those who’d like to use SQLite for a backend, running it as a separate daemon could be an interesting choice, which would also remove there need of Cgo for the build and maybe make things like separate background job processes easier to accomplish. See [1], [2].

—-

1: https://github.com/tursodatabase/libsql/tree/main/libsql-ser...

2: https://news.ycombinator.com/item?id=38602175

https://www.sqlite.org/copyright.html

To summarize, instead of using one of the OSS licenses, the copyright holders simply declare the source to be in the public domain. In order to preserve that status they don't accept patches unless you submit some signed document that you agree with that.

To make things more complicated, they also use their a relatively niche version management system instead of git. Which would complicate making contributions (if they accepted them).

There's a popular fork that fixes all of these issues: https://github.com/libsql/libsql It is MIT licensed, on Github, and open for contributions.

Kind of a weird legal situation for a popular project like this that so many people depend on to have. Not judging; but it is odd. Seems like a lot of wasted efforts between users, would be contributors, and the people that forked this thing to address all that.

You could try making feature requests for https://github.com/libsql/libsql , which is a community fork of SQLite that aims to speed-up the development of long-wanted features.

For a comprehensive view, check out the issues list for libSQL core and sqld. But mostly, I want libSQL to be a home for all builders who believe there is room to take a lean, mean, and SQLite-compatible embedded database to new heights. I’d love to see your contribution!