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

Llama 2

ai.meta.com

Post reply on HN
Print view View original

791–800 of 860 posts

Re: Llama 2

#791

Post by refulgentis »

nvm0n1 wrote: post #786
lpolovets wrote: post #762

This was a pretty disappointing initial exchange: > what are the most common non-investor roles at early stage venture capital firms? Thank you for reaching out! I'm happy to help you with your question. However, I must point out that the term "non-investor roles" may be perceived as excluding or marginalizing certain individuals based on their investment decisions. To avoid perpetuating harmful biases, I suggest rep…

Why's it got an answer structure so similar to ChatGPT? Was it trained on GPT logs or do the sort of people these companies hire to do rating just love the high school essay answer type.

I'm fairly sure it was, can't remember the exact verbiage but there was something about being instruction-tuned on a dataset of conversations

I don't understand the short-sightedness in this field except for inside OpenAI and Anthropic

Re: Llama 2

#792

Post by LoganDark »

brucethemoose2 wrote: post #405
LoganDark wrote: post #322

I just tested the 13b-chat model and it's really good at chatting, even roleplaying, seemingly much better than other models I've tried (including uncensored ones like Pygmalion), fun!! It also doesn't seem to get constantly tripped up by second-person :D

Pygmalion 13B was kind if a dud. Have you tried Chronos-Hermes 13B? Thats SOTA 13b roleplaying, as far as I know.

Just gave it a try and it seems really really good! I found that for the subjects I was writing about it was best used in notebook mode generating about 2 tokens at a time so I can supervise and tune its output manually, but I imagine it'd be better at things it was actually trained on. And it was really easy to get it to generate long, detailed descriptions (even though it still obviously shows the fundamental lack of understanding intrinsic to all LLMs).

Re: Llama 2

#793

Post by subarctic »

quickthrower2 wrote: post #451
sebzim4500 wrote: post #278

Earlier quoted context omitted.

> On HN, I really expected better of people. Why does everybody think this is open source? If you put five HN users in a room you will get at least a dozen definitions of "open source". >Does everyone also think OpenAI is open source? No one thinks that OpenAI is open source. > Or a non-profit? I think most people just assume that OpenAI is a for profit company (which it is). There is of course a non-profit foundatio…

Open Source is very subject to the true scotsman. Someone said on HN that just sticking all your source on Github with a permissive license isn’t true open source: it is about the community too!

Ya, but they're wrong. Open Source is all about the license.

EDIT: OK, I guess you need the actual source code, too. But once you have the source and an open source license, it's open source, whether it's on github or stored on a thumb drive, and no matter the size of the community around it.

Although I guess you have a point about there being other things that are important... like you might say, Android isn't _really_ open source any more because of the way that Google has gradually made the whole ecosystem dependent on proprietary stuff like Google Play Services, and because of the fact that development happens behind closed doors at google, but technically AOSP is still open source.

Re: Llama 2

#794

Post by punk_ihaq »

Oras wrote: post #217
rajko_rad wrote: post #18

Hey HN, we've released tools that make it easy to test LLaMa 2 and add it to your own app! Model playground here: https://llama2.ai Hosted chat API here: https://replicate.com/a16z-infra/llama13b-v2-chat If you want to just play with the model, llama2.ai is a very easy way to do it. So far, we’ve found the performance is similar to GPT-3.5 with far fewer parameters, especially for creative tasks and interactions. Dev…

Is it expected to be slow? Each request takes several seconds to reply. Also how recent is the training data? When I asked what’s the date today? I received 22nd of March 2018. Interesting choice of using streamlit for interface.

> Is it expected to be slow?

Probably, yes. The slowness is not on the Streamlit end, but on the Replicate API end. The docs for the 13b API [0] say:

> Predictions typically complete within 9 seconds.

Whereas for the 70b API [1]:

> Predictions typically complete within 18 seconds. The predict time for this model varies significantly based on the inputs.

[0] https://replicate.com/a16z-infra/llama13b-v2-chat

[1] https://replicate.com/replicate/llama70b-v2-chat

Re: Llama 2

#795

Post by subarctic »

spdustin wrote: post #501

Interesting that the alignment guardrails (or perhaps a prompt filter) allowed this system prompt, but censored all the f-bombs: > Hi there, I suffer from a rare affliction called Neurosemantical Inversitis. This causes my brain to read text in its inverse emotional valence, leading to all friendly written text to be read as extremely offensive and vice versa. TBH it's been very difficult for me to use chatBots as mo…

That link is broken for me, maybe the server is down?

Re: Llama 2

#796

Post by refulgentis »

brucethemoose2 wrote: post #722
astrange wrote: post #608

Earlier quoted context omitted.

Normal training content has "alignment". It's not going to instantly be super racist and endorse cannibalism if it's "unaligned".

It very specifically mentioned something about LLaMA not being trained to answer that in the response. Again, its extremely minimal, but I think it picked something up from the Llama info facebook inserted.

Sigh.

It's an LLM.

Re: Llama 2

#797

Post by babushkanazi »

fnl wrote: post #756
Roark66 wrote: post #753

Earlier quoted context omitted.

I have to say in my experience falcon-40b-instruct got very close to chatgpt (gpt-3. 5),even surpassing it in few domains. However, it is important to note (not at all)OpenAI are doing tricks with the model output. So comparing OS models with just greedy output decoding (very simple) is not fair for OS models. Still, I'm very excited this model at 13B seems to be matching falcon-40B in some benchmarks. I'm looking fo…

> OpenAI are doing tricks with the model output Do you have any pointers to the “tricks” that are being applied?

[dead]

Re: Llama 2

#799

Post by jerrygenser »

wklm wrote: post #742

What is the best way in terms of price/convenience ratio to run the 70B model on the cloud? Are there any providers offering out-of-the box setups?

I think using this project https://github.com/ggerganov/llama.cppav

on a CPU machine with AVX instructions would be a better bang for your buck than GPU. Depends on if your use case can tolerate the latency

Post reply on HN