Fine-tuning is a much better proposition than you’re giving it credit for. Papers are coming out demonstrating that 7B parameter models can outperform GPT-4’s quality when trained on a limited set of tasks. Yet, a 7B model offers comparatively cheap and fast inference. Furthermore, for a lot of use cases, few-shot priority is infeasible because you need to supply 2-3k tokens worth of few-shot examples with every prompt in order to fully specify the behavior you want. (As an example, think of long-form summarization where you want the summary to adhere to certain rules.)

What kind of tasks? Could you give some links to the papers you're referring to?

Sure:

1. Goat: Fine-tuned LLaMA Outperforms GPT-4 on Arithmetic Tasks

https://huggingface.co/papers/2305.14201

2. Gorilla: Large Language Model Connected with Massive APIs

https://arxiv.org/abs/2305.15334

Sorry I should have been more specific - I was limiting my question to the bigger models. The smaller (~7B) models are feasible with these approaches.

Ah. Perhaps the larger models will find use in in-house deployments where companies want their employees to have access to ChatGPT-like general purpose assistants, but want to prevent data from leaving their premises. LIMA shows the LLaMA 65B hitting a quality level somewhere between DaVinci-003 and GPT-4 with minimal alignment, so the base models are probably powerful enough already for this to work. Just speculating.

One benefit of finetuning larger models, like 65B, is to free up limited context space vs few-shot prompting.

If you want a specific kind of interaction with the model then you could take up 1/3rd of the 2048 token context window with few-shot or you could simply finetune it with QLoRA for a few hours on a consumer GPU and then get to use the full 2048 context with the finetuned model.

Wow, thanks a lot! :)

I'm running Vicuna on a free 4core Oracle VPS, and it's perfectly usable for a Discord bot. Responses rarely take more than 15 seconds with <256 max token limit, and the responses are much more entertaining than GPT 3.5. I'm not using the streaming API my server software[0] offers, but if I did it would probably load somewhere between the speeds of GPT-3.5 and GPT-4. It's more or less the same time a human would take to compose the same message.

So... not exactly a serious use-case. But it's what I'm using, and now I'm saving 10s of dollars on inferencing costs per month!

[0] https://github.com/go-skynet/LocalAI

Now I’m curious what your bot does!

It's one of those "say a keyword with a question, get a response" type bots. I added in a few other "prompt sources" though, where it grabs the first part of an RSS entry or HN comment and tries to autocomplete the rest. Mostly just a boring testbed for me to play with models, for free, with friends.

TIL Oracle has VPS offerings with a free tier. Are they any good? Is the free-tier time limited?

This use-case is alright for a toy I guess - which is the extent that I was originally expecting these things to be useful for.

They're okay. This isn't the place for a full review of their offerings (especially considering everyone's mixed feelings on Oracle), but I'm confident that it's better than most 1core/$5 deals you'll find elsewhere.

> Are they any good?

Yep, free tier allows you to spec up to 24gb of RAM without paying, which is cool. The bottleneck is really the disk speed, but that's not an issue with mmaped models. There's enough cached memory that it loads instantly, so it's good-ish for this use case.

> Is the free-tier time limited?

No, but there are a lot of strings attached:

- The cores are vCPUs, not dedi (duh)

- You can't create new instances when demand is high (unless you add a credit card)

- Technically Oracle reserves the right to shut down the instance if demand gets really high (although I haven't heard any stories about this personally)

Proceed with caution. It's still a great place to start before you shell out $1/hr for dedi GPU rackspace.

They will shut down the VPS if there's no activity on it, not sure how they detect this though

CPU idle % over a rolling time window (say, 15 minutes).

>From what I understand, even to run locally you/your team needs to be able to afford a machine with a 4090. These are super expensive in some countries.

That's because we're only half a year into LLMs becoming mainstream. Give it 3-4 years. The advancements in bringing down model size, optimizations, and newer GPUs, SoCs from Nvidia, AMD, Apple, Intel, Qualcomm, etc will make it so that top LLMs will run on a highend laptop/desktop.

This is bleeding edge stuff.

All advances in this direction do indicate that it will be easier and easier for more people to do things with it.

This doesn't need to work for everyone.

A 4090 costs today 2k, the 3090 with also 24gb costs today 1k and costed 2k.

Any particular reason not to run this model on a single Jetson AGX Orin 64GB?

GP Core count is much lower than than the 4090 but it still does 275 int8 TOPS for only $2k

I'm the wrong person to ask but performance wise a 4090 has over a pflop and a Google search also showed a factor of 3-4 for a 3090.

The relevant performance metric here is memory amount and bandwidth.