WizardLM VS guidance

This is interesting work, and a good contribution, but there is no need to mislead people.

[1] https://github.com/nlpxucan/WizardLM

If you pair this with the latest WizardCoder models, which have a fairly better performance than the standard Salesforce Codegen2 and Codegen2.5, you have a pretty solid alternative to GitHub Copilot that runs completely locally.

(We will update the demo links in our github.)

In terms of speed, we're talking about 140t/s for 7B models, and 40t/s for 33B models on a 3090/4090 now.[1] (1 token ~= 0.75 word) It's quite zippy. llama.cpp performs close on Nvidia GPUs now (but they don't have a handy chart) and you can get decent performance on 13B models on M1/M2 Macs.

You can take a look at a list of evals here: https://llm-tracker.info/books/evals/page/list-of-evals - for general usage, I think home-rolled evals like llm-jeopardy [2] and local-llm-comparison [3] by hobbyists are more useful than most of the benchmark rankings.

That being said, personally I mostly use GPT-4 for code assistance to that's what I'm most interested in, and the latest code assistants are scoring quite well: https://github.com/abacaj/code-eval - a recent replit-3b fine tune the human-eval results for open models (as a point of reference, GPT-3.5 gets 60.4 on pass@1 and 68.9 on pass@10 [4]) - I've only just started playing around with it since replit model tooling is not as good as llamas (doc here: https://llm-tracker.info/books/howto-guides/page/replit-mode...).

I'm interested in potentially applying reflexion or some of the other techniques that have been tried to even further increase coding abilities. (InterCode in particular has caught my eye https://intercode-benchmark.github.io/)

[1] https://github.com/turboderp/exllama#results-so-far

[2] https://github.com/aigoopy/llm-jeopardy

[3] https://github.com/Troyanovsky/Local-LLM-comparison/tree/mai...

[4] https://github.com/nlpxucan/WizardLM/tree/main/WizardCoder

You talking about this? https://github.com/nlpxucan/WizardLM

The smallest model that is close to competent at code is WizardCoder 15B.. https://github.com/nlpxucan/WizardLM/

WizardCoder: Empowering Code Large Language Models with Evol-Instruct (https://github.com/nlpxucan/WizardLM/tree/main/WizardCoder)

This IS Microsoft Guidance, they seem to have spun off a separate GitHub organization for it.

https://github.com/microsoft/guidance redirects to https://github.com/guidance-ai/guidance now.

... and it sets the value of "armor" to "leather" so that you can use that value later in your code if you wish to. Guidance is pretty powerful, but I find the grammar hard to work with. I think the idea of being able to upload a bit of code or a context-free grammar to guide the model is super smart.

https://github.com/microsoft/guidance/blob/d2c5e3cbb730e337b...

Here's one thing I don't get.

Why all the rigamarole of hoping you get a valid response, adding last-mile validators to detect invalid responses, trying to beg the model to pretty please give me the syntax I'm asking for...

...when you can guarantee a valid JSON syntax by only sampling tokens that are valid? Instead of greedily picking the highest-scoring token every time, you select the highest-scoring token that conforms to the requested format.

This is what Guidance does already, also from Microsoft: https://github.com/microsoft/guidance

But OpenAI apparently does not expose the full scores of all tokens, it only exposes the highest-scoring token. Which is so odd, because if you run models locally, using Guidance is trivial, and you can guarantee your json is correct every time. It's faster to generate, too!

Perhaps something as simple as stating it was first built around OpenAI models and later expanded to local via plugins?

I've been meaning to ask you, have you seen/used MS Guidance[0] 'language' at all? I don't know if it's the right abstraction to interface as a plugin with what you've got in llm cli but there's a lot about Guidance that seems incredibly useful to local inference [token healing and acceleration especially].

[0]https://github.com/microsoft/guidance

LangChain is just too much, personal solutions are great, until you need to compare metrics or methodologies of prompt generation. Then the onus is on these n-parties who are sharing their resources to ensure that all of them used the same templates, they were generated the same way, with the only diff being the models these prompts were run on.

So maybe a simpler library like Microsoft's Guidance (https://github.com/microsoft/guidance)? It does this really well.

I am unclear on the status of the project but here is the conversation that seem to be tracking it: https://github.com/microsoft/guidance/discussions/201

Let's have a subreddit about Guidance!

You should look at guidance for an LLM to fill out a template. Define the output data structure and provide the real estate listing in the context (see the JSON template example here https://github.com/microsoft/guidance)