Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Alpaca: A strong open-source instruction-following model

crfm.stanford.edu

Post reply on HN
Print view View original

71–80 of 313 posts

Re: Alpaca: A strong open-source instruction-following model

#71

Post by nickthegreek »

Traubenfuchs wrote: post #42
margorczynski wrote: post #17

> Importantly, we have not yet fine-tuned the Alpaca model to be safe and harmless. We thus encourage users to be cautious when interacting with Alpaca, and to report any concerning behavior to help improve the safety and ethical considerations of the model. Thanks but I would prefer you don't start injecting your "safeness" to gimp it. Or at least provide it as an alternative.

[flagged]

It makes sense if you are selling the API to a bunch of businesses that will be building it into customer facing applications. Ideally though, it would be a toggle that the developer could switch based on their usecases.

Re: Alpaca: A strong open-source instruction-following model

#72

Post by dmw_ng »

2bitencryption wrote: post #46
simonw wrote: post #28

This is why I think we're seeing a Stable Diffusion moment for LLMs: https://simonwillison.net/2023/Mar/11/llama/ Look at the timeline: 24th February 2023: LLaMA is announced, starts being shared with academic partners: https://research.facebook.com/publications/llama-open-and-ef... 2nd March: Someone posts a PR with a BitTorrent link to the models: https://github.com/facebookresearch/llama/pull/73 10th March: First…

Here is one question I have not seen answered yet: All the magic of "7B LLaMA running on a potato" seems to involve lowering precision down to f16 and then further quantizing to int4. Clearly this quantized model still outputs something resembling human language, at the very least. But I haven't seen anyone show what effect this quantizing has on the quality of the output. If the quality of the output is bad, it's un…

For 10 billion+ parameter models, the effects of quantization are relatively small, for smaller models like Llama 7B the effect becomes more dramatic, but there is ongoing research on new quantization methods (like GPTQ) that preserve significant performance even on the lower end.

Quantization isn't the only technique available for downsizing a model, Llama itself is already the result of sizing the model and input data according to "Chinchilla optimality", a very recent (as in 2022) result that e.g. GPT-3 predates. The result is that Llama-13B performs in benchmarks similarly with GPT-3 175B despite the tremendous size difference. There are separately also a variety of pruning methods to further eliminate inactive weights present in the trained model (I think this is also active research)

Finally even on something like a Raspberry Pi, implementations for inference (like llama.cpp) are nowhere near mature yet. There are already a multitude of runtimes available for inference making large tradeoffs between performance and flexibility (e.g. many models running on PyTorch vs ONNX report 5-10x speedups running under ONNX)

I think the really exciting part of Alpaca is the size and budget of the team - 5 students with $100 scraping OpenAI put this model together in a couple of hours of training. Any notions of premium persisting in the AI space for much longer seem fantastic at best, for all intents and purposes it has already been commoditized. And that's scary considering the size of the dent ChatGPT has put in my Google traffic

Re: Alpaca: A strong open-source instruction-following model

#73

Post by byteknight »

gamegoblin wrote: post #62
2bitencryption wrote: post #46

Earlier quoted context omitted.

Here is one question I have not seen answered yet: All the magic of "7B LLaMA running on a potato" seems to involve lowering precision down to f16 and then further quantizing to int4. Clearly this quantized model still outputs something resembling human language, at the very least. But I haven't seen anyone show what effect this quantizing has on the quality of the output. If the quality of the output is bad, it's un…

Decrease in accuracy is negligible and decreases as model size increases. That is, larger models quantize even better than smaller models. https://arxiv.org/abs/2210.17323

Is this because averages are weighed less (less sensitive) as the total sample size increases?

Re: Alpaca: A strong open-source instruction-following model

#74

Post by Traubenfuchs »

beepbooptheory wrote: post #65
Traubenfuchs wrote: post #51

Earlier quoted context omitted.

[flagged]

Its sarcastic but how are you literally not that person right now? Being angry at an invisible evil, not once opening your eyes to any other thought, totally incapable of making any real point except to whine... It's quite a good descriptor actually! I hope you can find a model for yourself that will assure you there are only two genders. Just to save the rest us from this childishness.

That's like telling someone who uses violence in self defense that he is now just as bad as the attackers he defended himself against. Like telling someone who shouts at people to turn down their music that, gotcha, now you were shouting too.

Re: Alpaca: A strong open-source instruction-following model

#75

Post by leobg »

londons_explore wrote: post #23

The training data doesn't seem to be great quality... "instruction": "Calculate the sum of each column in the following table.", "input": "2 | 3 | 4\n 5 | 6 | 7", "output": "2+3+4 = 9\n5+6+7 = 18" I think better results could be obtained by manually going through these 52,000 training examples - even a couple of seconds per example should be enough to weed out obviously wrong ones, and should only take <$1000 of huma…

Don’t even need human time. Just run it through davinci. Is this correct, yes or no? Or MTurk. Can cost less than $500.

Re: Alpaca: A strong open-source instruction-following model

#76

Post by margorczynski »

soVeryTired wrote: post #70
margorczynski wrote: post #63

Earlier quoted context omitted.

> If we can't get models not to say racist or otherwise terrible things, we can't make any guarantees about our ability to control or guide some future AGI. This is a very bold assumption that the current LLMs function and "think" in the same way some future AGI would. They do not even reason, just make up words that fit some context - thus they "hallucinate". There is no reason the approach taken here by injecting s…

> They do not even reason, just make up words that fit some context - thus they "hallucinate". But they can explain their 'reasoning' in a way that makes sense to humans a lot of the time. Serious question: how do you know if something does or doesn't reason?

That is not their reasoning though - it is something they think a human would write given the predicate (question that expects you to provide the reasoning behind the answer). For something to reason it needs the ability to have certain goals and to perform action which it thinks are the most optimal to reaching those goals. Like setting hypotheses and producing a path towards proving them - reasoning.

The LLM only correlates, so it's "reasoning" is something like "most often people answered 4 to 2+2 then that I should write". That's why it gives out confidently complete gibberish as it works with correlation and not causality. I think much closer to that goal of real reasoning are world models - check out something like DreamerV3 or what Yann Le Cunn is talking about.

Re: Alpaca: A strong open-source instruction-following model

#78

Post by ben_w »

Traubenfuchs wrote: post #42
margorczynski wrote: post #17

> Importantly, we have not yet fine-tuned the Alpaca model to be safe and harmless. We thus encourage users to be cautious when interacting with Alpaca, and to report any concerning behavior to help improve the safety and ethical considerations of the model. Thanks but I would prefer you don't start injecting your "safeness" to gimp it. Or at least provide it as an alternative.

[flagged]

If your takeaway from…

> Importantly, we have not yet fine-tuned the Alpaca model to be safe and harmless.

…is "oh no I can't get it to emit amusing racial and sexual slurs", you've not understood the problem of AI safety.

This is not why US broadcast television can have people say they've pricked their finger but not vice versa.

It is the entire history of all the controversies of The Anarchist Cookbook, combined with all the controversies about quack medicine, including all the ones where the advocates firmly believed their BS like my mum's faith in Bach flower and homeopathic remedies[0]; combined with all the problems of idiots blindly piping the output to `exec`, or writing code with it that they trust because they don't have any senior devs around to sanity check it because devs are expensive, or the same but contracts and lawyers…

And that's ignoring any malicious uses, though fortunately for all of us this is presently somewhat too expensive to be a fully-personalised cyber-Goebbels for each and every sadistic machiavellian sociopath that hates you (the reader) personally.

[0] which she took regularly for memory; she got Alzheimer's 15 years younger than her mother who never once showed me any such belief.

Re: Alpaca: A strong open-source instruction-following model

#80

Post by nlowell »

beepbooptheory wrote: post #52
Traubenfuchs wrote: post #42

Earlier quoted context omitted.

[flagged]

Its really just because of boring old liability, nothing more. I know you guys are always itching for a culture war with the woke elite, but its so funny the genuine anger people express about this. Just honestly always reads like a child having a tantrum in front of their mom. Can't yall like pick on the opinions of teenagers like you normally do? This very project shows you can make your own AI as edgy as you want…

You're missing the point and willfully characterizing others as solely being concerned with making the AI's say slurs. That's not their concern. But you can win any imaginary argument you like.
Post reply on HN