outlines VS hof

> [2]: https://github.com/outlines-dev/outlines?tab=readme-ov-file#...

It's interesting as speech recognition has become more popular than ever through services like Alexa, and other iot devices support for OS speech recognition

Unfortunately most implementations (especially those that are iot focused) don't have very important features for robust speech recognition.

1. Ability to enable and disable a grammar

[4] https://github.com/outlines-dev/outlines/tree/main

The next step is defining how to guide generation. For this step, we'll use the Outlines library. Outlines is a library for controlling how tokens are generated. It applies logic to enforce schemas, regular expressions and/or specific output formats such as JSON.

No benchmarks, just my anecdotal experience trying to get local LLM's to respond with JSON. The method above works for my use case nearly 100% of the time. Other things I've tried (e.g. `outlines`[0]) are really slow or don't work at all. Would love to hear what others have tried!

0 - https://github.com/outlines-dev/outlines

As I was playing with the Outlines library (https://outlines-dev.github.io/outlines/), I discussed with my friend Maxime how funny it would be if we set up a way to pair LLMs in chess matches till one wins. The first time I tried it, it required substantial prompt engineering to get some of those LLMs to propose valid moves. Large language models can mostly stay focused and even play rather well; see https://news.ycombinator.com/item?id=37616170 for example. However small language models aren't as easy to convince.

Some of those LLMs have seen very little chess notation and so after the first few opening moves there aren't any valid tactics, let alone strategy, so they would end up either repeating the same move, or hallucinate moves that are not valid (Kxe5, but there would be a queen on e5!)

Then Outlines came along and we could force them to pick valid moves with little cost! Maxime worked super fast and got a first version of this idea as a gradio space.

I think it is pretty fun to see the (mostly terrible, but otherwise valid) chess that those LLMs play. Maybe it will even be instructive to how we can create small LLMs that can play much better than the ones on the leaderboard.

Anyway, you can check it out here:

https://huggingface.co/spaces/mlabonne/chessllm

What is interactive about it: you can pick the LLMs from available models on HuggingFace (within reason, small LLMs are preferable so that the space does not crash) or push one of your own small models to HF and have it fight with others. At the end of the game the leaderboard is updated.

Hope you find it fun!

> The most obvious usage of this is forcing a model to output valid JSON

Isn't this something that Outlines [0], Guidance [1] and others [2] already solve much more elegantly?

0. https://github.com/outlines-dev/outlines

1. https://github.com/guidance-ai/guidance

2. https://github.com/sgl-project/sglang

And I think “just” Markdown doesn’t quite cut it for safe guidance. For example: directly generating content for your components. But I’m really excited about tooling like outlines appearing, with a greater focus on guided generation for structured data. Because this is often what we actually need!

It's pretty easy to force a locally running model to always output valid JSON: when it gives you probabilities for the next tokens, discard all tokens that would result in invalid JSON at that point (basically reverse parsing), and then apply the usual techniques to pick the completion only from the remaining tokens. You can even validate against a JSON schema that way, so long as it is simple enough.

There are a bunch of libraries for this already, e.g.: https://github.com/outlines-dev/outlines

https://github.com/outlines-dev/outlines/blob/7fae436345e621... squares with my experience using LLMs for anything real

  sequence = generator("Alice had 4 apples and Bob ate 2. Write an expression for Alice's apples:")

For a auth product, you probably should go with traditional code generation, not LLMs (re: all the unsolvable problems they have (hallucinations, incomplete code answers). For an auth product, you want something definite and reproducible, or you are asking for trouble.

If you are interested in a flexible code gen framework, I built one that had this very use case in mind. (https://github.com/hofstadter-io/hof)

I'd be happy to think / talk this through with you

I recently built a TUI for exploring and developing CUE, JSON, Yaml.

Large inspiration for the overall UX comes from the Bloomberg terminal, where every function is reachable with four letter shortcut from the command box. It was really the command box that I liked. I've also built a sort of "flex" panel component for the layout so you can create as many panels as you like.

https://docs.hofstadter.io/getting-started/hof-tui/

https://github.com/hofstadter-io/hof/tree/_dev/lib/tui

Many business analysts use SQL, have for a long time. They are probably not your target audience. With the problem being JAVA specific, you'd likely want to start there

This sounds similar to the goals of my hof tool (https://github.com/hofstadter-io/hof), lift type definitions out of code so they can be defined in one place, then generate the code for all the places. Is that sounding like what you are after?

Having built something similar, the biggest challenge for users is that they have to use a bespoke language, like WASP here. I suspect that it is also your biggest challenge as well.

Mine is built on CUE, which at least has the potential to become a more widely used language. CUE hasn't reached sufficient maturity for broader adoption yet, so I continue to face this same problem.

https://github.com/hofstadter-io/hof

Here's a big one I needed to get ChatGPT to do something more sophisticated with a JSON object response (predates functions and all that)

https://github.com/hofstadter-io/hof/blob/_dev/flow/chat/pro...

It no longer worked after a model update some time ago, haven't tried recently.

I found codellama to be much better for this and require fewer instructions, an anecdotal validation for smaller, focussed models

I was surprised out how you can define arbitrary grammars using arbitrary formulation and it would follow it. Of course you have to redo the prompt every time there is an update... such a pain

https://github.com/hofstadter-io/hof/blob/_dev/flow/chat/pro...

CUE is indeed a beautiful language, will get those mind juices flowing for sure!

There is more work to be done on the codec implementation, but if you just want to split yaml/json across files, CUE is a great option

You might also like my project, built on CUE: https://github.com/hofstadter-io/hof We have a TUI where you can explore and work with CUE, JSON, Yaml

You can pretty much make up any pseudo grammar like this one, which is a reduced JSON object that is close to CUE: https://github.com/hofstadter-io/hof/blob/_dev/flow/chat/pro...

No need to be formal or use a standard format

Yea, in particular for this project, they have created a bespoke templating system.

You can get the same thing with Go text/templates by adding chat function(s) as custom a helper: https://github.com/hofstadter-io/hof/blob/_dev/lib/templates...

100% one of the best things about building a TUI is not having the pain of modern web development. I do think there is a way to have a CLI & TUI come from the same code, so you can get the best of both, or pick the best for the task at hand.

experiments in progress here: https://github.com/hofstadter-io/hof/tree/_dev/lib/tui