Sail 7B: New Fine Tuned LLM Outperforms ChatGPT and Vicuna with Search | Hacker News Reader

Sail 7B: New Fine Tuned LLM Outperforms ChatGPT and Vicuna with Search

openlsr.org

81 pointsyujian17 comments

Demo isn't really connected to the model, no way to test it out. Last commit was two weeks ago.

Did anyone reproduce the outlined steps? If this thing can be stronger than GPT-3.5 at 7B parameters (which I find unlikely), it would make bigger splash than it did.

Sry our backend gpu servers are being upgraded and it caused some problem for our job submission system.

Our slurm is not fixed yet, but I've brought the model back online for now - it will be temporally diabled again when our slurm is fixed.

Gradio does not allow tasks more than a minute, so the demo might fail for long texts.

A bit misleading to say ChatGPT, when GPT-4 according to the metrics there seem to outpeform it?

At this point a lot of people use "ChatGPT" to refer to the ChatGPT model with the GPT-3.5 backend

To me it's a collection of both, so the title makes it sound that it performs better than GPT-4. I don't know who those people are exactly who only hold GPT-3.5 to that meaning.

I imagine the paid users are a minority, so almost everyone

Even in Hackerrank and amongst technical people whom the article I would imagine is directed to?

Probably the free users.

And a lot of people don't. In particular, all the stories in the media about ChatGPT's unexpectedly good performance at various tasks, are more likely than not to be about GPT-4 model.

This is a legitimate point of confusion, and it's best to keep it explicit which model is being discussed. Really should be written at the start of the text.

Thanks! will replace "chatgpt" with "gpt-3.5-turbo".

https://github.com/luohongyin/SAIL

> outperforms ChatGPT

Doubt.

I mean, maybe in a while, but it seems like the general trend is that a month or two after these announcements, independent researchers end up finding that the opposite is actually true.

The model outputs ChatGPT by connecting to a search engine - it's expected because it is fed with search results. Without search, SAIL-7B does not outperform ChatGPT.

The interesting aspect is that the ChatGPT response quality is not improved with search results, and it can be misled by noisy search.

Another model that isn't actually able to be used commercially because of links to LLaMa base weights.

It is a problem, but the same training set can be used to train commercial models.

The reddit link is just an empty page with another link. It would be better to use https://openlsr.org/sail-7b

Changed to that from https://old.reddit.com/r/AI_Agents/comments/140e03v/sail_7b_.... Thanks!

Reply on news.ycombinator.com