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

151–160 of 226 posts

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

#151

Post by aqader »

adultSwim wrote: post #99

The steps to recreate Alpaca are openly available. Is anyone posting the resulting model? (LLaMA fine-tuned to follow instructions)

There are a couple open source implementations. I'll list a couple below:

7B: - https://huggingface.co/tloen/alpaca-lora-7b - https://huggingface.co/ozcur/alpaca-native-4bit 13B: - https://huggingface.co/samwit/alpaca13B-lora - https://huggingface.co/Dogge/alpaca-13b 30B: - https://huggingface.co/baseten/alpaca-30b - https://huggingface.co/Pi3141/alpaca-30B-ggml

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

#152

Post by NERD_ALERT »

earnesti wrote: post #53
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 doesn't have an identity, so it just creates something random. It is irrelevant for many use cases I would guess.

Hmm I'm not sure here. I asked "Who are you?" and first I got this response.

  LLaMa didn’t return anything, please try another prompt. It 
  responds best to questions.
Then when I asked again I got the same response as the parent comment, word for word.

  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.

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

#153

Post by tjr »

nathias wrote: post #131

> me: what does Hegel think about negativity? > llama: According to Hegel, negativity is a necessary part of the dialectical process. He believed that negativity is the driving force of history, and that it is necessary in order for progress to be made. Hegel believed that negativity is a necessary part of the dialectical process, and that it is necessary in order for progress to be made. Hegel believed that negativi…

> me: write a y combinator in C

> llama:

  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include 
  #include
> me: write a Y Combinator in Common Lisp

> llama:

  (defun y-combinator (x y)
  (if (null? x) 
  y
  (cons (car x) (y-combinator (cdr x) y))))

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

#155

Post by gorbypark »

nico wrote: post #105
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’m sure someone will eventually pony up for the compute time to do 65B. How much would it cost? Can you give a breakdown of hardware/paas requirements and the costs?

Another datapoint, first hand report from a random internet person that 7B took 2.5-3hrs on a 8xA100 80Gb setup. If it ends up being a straight up linear thing, that means about ~27 hours for the 65B model. Depending on the host and if it's a preemptable instance or not, that could be about $12-30 per hour.

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

#156

Post by nullstyle »

gorbypark wrote: post #155
nico wrote: post #105

Earlier quoted context omitted.

> I’m sure someone will eventually pony up for the compute time to do 65B. How much would it cost? Can you give a breakdown of hardware/paas requirements and the costs?

Another datapoint, first hand report from a random internet person that 7B took 2.5-3hrs on a 8xA100 80Gb setup. If it ends up being a straight up linear thing, that means about ~27 hours for the 65B model. Depending on the host and if it's a preemptable instance or not, that could be about $12-30 per hour.

Have a link? I haven't seen any finetuning scripts in the wild that train a PEFT model on a multrigpu setup yet and would love to play around with one.

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

#159

Post by gorbypark »

nullstyle wrote: post #156
gorbypark wrote: post #155

Earlier quoted context omitted.

Another datapoint, first hand report from a random internet person that 7B took 2.5-3hrs on a 8xA100 80Gb setup. If it ends up being a straight up linear thing, that means about ~27 hours for the 65B model. Depending on the host and if it's a preemptable instance or not, that could be about $12-30 per hour.

Have a link? I haven't seen any finetuning scripts in the wild that train a PEFT model on a multrigpu setup yet and would love to play around with one.

The original Alpaca repo has the training script. The readme has the torchrun command and arguments used for train.py. https://github.com/tatsu-lab/stanford_alpaca/blob/main/train...

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

#160

Post by moffkalast »

dmd wrote: post #44

Me: How can I convince my 5 year old to put her shoes on in a timely fashion each morning? Answer in table form. LLaMA: 1. 2. 3. 4. 5. 1. 2. 3. 4. 5. 1. 2. 3. 4. 5. 1. 2. 3. 4. 5. 1. 2. 3. 4. 5. 1. 2. 3. 4. 5. 1. 2. 3. 4. 5. 1. 2. 3. 4. 5. 1. 2. 3. 4. 5. 1. 2. 3. 4. 5. 1. 2. 3. 4. 5. (gpt: https://cloud.typingmind.com/share/0841633b-5150-4f7c-a370-0... )

The performance is surprisingly similar to one of those FLAN models. Can't exactly expect much from the 7B version but the 30B one ought to be interesting once it's up.
Post reply on HN