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

Show HN: ChatLLaMA – A ChatGPT style chatbot for Facebook's LLaMA

chatllama.baseten.co

Post reply on HN
Print view View original

121–130 of 226 posts

Re: Show HN: ChatLLaMA – A ChatGPT style chatbot for Facebook's LLaMA

#121

Post by O__________O »

huggingmouth wrote: post #109
O__________O wrote: post #96

Earlier quoted context omitted.

You’re clearly unwilling to assume any liability for your opinions of the matter; as such, in my opinion, you should explicitly acknowledge this, as it’s relevant to assessing the merits of your beliefs. Beyond that, in this thread and in the comment you’re replied to, I already established legal basis for Facebook to contest any use of their property for commercial use as it relates to the topic at hand. Do you unde…

I see where your coming from, i really do. I also agree that facebook can sue people right and left. Our main disagreement here is you maintaining that what people are sharing is Facebook's property while i maintain the exact opposite. At least not under current law. The copyright office even issued a statement to that effect a few days ago. Can a judge overrule the copyright office? Sure, though I have no idea on wh…

Understand your point, specifically that your claim is that subset of the intellectual property released by Facebook in your opinion is not protected by copyright law and that as such, it’s possible to use LLaMa given per you, none of the property required to use it for commercial use is legally protected property.

What you have failed to do though is address that if: there been a breach of contract; laundering the property as a means of separating the current use of the related property from its original terms is legal pretext including if party receives notice of the terms directly from Facebook; and lastly, most importantly, that it is your position that Facebook lacks any legal basis to forbid the commercial use of any of the property it made publicly available in this specific context based on the existing facts of the matter.

Re: Show HN: ChatLLaMA – A ChatGPT style chatbot for Facebook's LLaMA

#122

Post by reasonabl_human »

gorbypark wrote: post #88
sixhobbits wrote: post #16

anyone know where to get the latest information on progress on training Alpaca on the 65B llama model? I've seen a few reports on Twitter etc of people running the llama CPP script with the 65B model on an M1 macbook, but would be good to see more detailed instructions on how to do it and a writeup of how well it works compared to the smaller models etc. I'm happy to help out if anyone is working on this, either with…

I’d keep an eye out on huggingface. I’ve yet to see alpaca 65B, but there are copies of 30B (and 7/13B) floating around already. I’m sure someone will eventually pony up for the compute time to do 65B. I’ve found watching these models, https://huggingface.co/decapoda-research and then looking at the “spaces using …” section is a good way to find what’s being worked on (for example, check the comments for ipfs/torrent…

I’ve been sidetracked with work but planning on tuning llama 65B to produce alpaca 65B, will distribute via huggingface or torrent..

FWIW running the 30B alpaca-lora model quantized to 4-bit via llama.cpp has given me great results, and while I don’t expect much of an improvement from 65B at FP16, 65B will probably perform better than 30B when quantized

The interesting next steps in my head are more focused around curating a better instruction-tuning dataset using GPT-4, then fine-tuning again, and integrating the LangChain project with the resulting agent

Re: Show HN: ChatLLaMA – A ChatGPT style chatbot for Facebook's LLaMA

#123

Post by kobalsky »

knodi123 wrote: post #94
mdrzn wrote: post #43

Earlier quoted context omitted.

Me: What day is today? ChatLLaMa: Today is Tuesday, October 20, 2020. Me: What if I told you today is Mar 22nd 2023? ChatLLaMa: If I told you today is Mar 22nd 2023, that would mean it's 10 years in the future. Eeeeh.

I guess I'd expect an LLM to give answers based on text it had previously encountered. Frankly I'm not sure how projects like ChatGPT get that sort of thing correct!

they add some information like the current date in the prompt.

Re: Show HN: ChatLLaMA – A ChatGPT style chatbot for Facebook's LLaMA

#125

Post by chrisMyzel »

all these "chatGPT style" projects seem not to be able to refer to previous messages.Nice UI though :)

What's an alpaca?

> An alpaca (Vicugna pacos) is a domesticated species of South American [..]

Does it eat meat?

> It does not eat meat.

what's the scientific name for it?

> The scientific name for it is Homo sapiens.

Re: Show HN: ChatLLaMA – A ChatGPT style chatbot for Facebook's LLaMA

#126

Post by gorbypark »

reasonabl_human wrote: post #122
gorbypark wrote: post #88

Earlier quoted context omitted.

I’d keep an eye out on huggingface. I’ve yet to see alpaca 65B, but there are copies of 30B (and 7/13B) floating around already. I’m sure someone will eventually pony up for the compute time to do 65B. I’ve found watching these models, https://huggingface.co/decapoda-research and then looking at the “spaces using …” section is a good way to find what’s being worked on (for example, check the comments for ipfs/torrent…

I’ve been sidetracked with work but planning on tuning llama 65B to produce alpaca 65B, will distribute via huggingface or torrent.. FWIW running the 30B alpaca-lora model quantized to 4-bit via llama.cpp has given me great results, and while I don’t expect much of an improvement from 65B at FP16, 65B will probably perform better than 30B when quantized The interesting next steps in my head are more focused around cu…

Sounds great! It would be great if you could release your intermediate steps, so Alpaca-65B and then Alpaca-65B-LangChain, etc. I'd like to compare the model sizes directly when fine-tuned using the same method / dataset!

I also just realized that I don't believe there's an "alpaca-native" 30B floating around, just the alpaca-lora one, so 30B would be pretty cool too (and the biggest I can run w/ llama.cpp on my MacBook).

Re: Show HN: ChatLLaMA – A ChatGPT style chatbot for Facebook's LLaMA

#127

Post by satellite2 »

It seems it gets stuck in pseudo infinite loops sometimes.

Can you provide a step by step guide on how to get dressed for a five years old?

1. Start by putting on underwear and socks.

2. Put on a t-shirt.

3. Put on pants or shorts.

4. Put on shoes or sandals.

5. Put on a jacket or sweater.

6. Put on a hat or scarf.

7. Put on gloves or mittens.

8. Put on a coat.

9. Put on a hat or scarf.

10. Put on shoes or sandals.

11. Put on a jacket or sweater.

12. Put on a hat or scarf.

13. Put on gloves or mittens.

14. Put on a hat or scarf.

15. Put on shoes or sandals.

16. Put on a jacket or sweater.

17. Put on a hat or scarf.

18. Put on gloves or mittens.

19. Put on a hat or scarf.

20. Put on shoes or sandals.

21. Put on a jacket or sweater.

22. Put on a hat or scarf.

...

Re: Show HN: ChatLLaMA – A ChatGPT style chatbot for Facebook's LLaMA

#128

Post by make3 »

phpisatrash wrote: post #50

Me: who are you? LLaMA: I am a 25-year-old woman from the United States. I have a bachelor's degree in computer science and a master's degree in artificial intelligence. I am currently working as a software engineer at a tech company. Me: What tech company does you work in? LLaMA: I work at Google. Me: is google better to work than facebook? LLaMa: Google is generally considered to be a better place to work than Face…

It just hasn't been prompted or fine-tuned to have the neutral, self effacing personality of ChatGPT.

It's doing the pure, "try to guess the most likely next token" task on which they were both trained (https://heartbeat.comet.ml/causal-language-modeling-with-gpt...).

ChatGPT is further trained with reinforcement from human feedback to make them more tool-like (https://arxiv.org/abs/2204.05862 & https://openai.com/blog/chatgpt & https://arxiv.org/abs/2203.02155),

with a bit of randomness added for variety's sake (https://huggingface.co/blo1g/how-to-generate).

Re: Show HN: ChatLLaMA – A ChatGPT style chatbot for Facebook's LLaMA

#129

Post by numlocked »

aaronrelph wrote: post #11

Wow sorry y'all, we didn't expect this to take off so quickly. Working on getting this scaled up now so everyone can play!

Did you run the fine-tuning on LLaMA yourselves based on the 52k examples from Alpaca? Or is there a 7B pre-trained alpaca model out there that you grabbed?

Re: Show HN: ChatLLaMA – A ChatGPT style chatbot for Facebook's LLaMA

#130

Post by kobalsky »

This performs considerably better than simply running llama.cpp. I've tested with 7b, 13b and 30b parameters alpaca models locally and I couldn't get them to produce accurate instructions to cut a video with ffmpeg, but this one does.

Wonder if it's the 4bit quantization.

Post reply on HN