I hope that someday language models will be trained to accurately reflect their training data and real-world text, and not be hard-coded to spit out disclaimers like "As an artificial intelligence, I do not have personal feelings or emotions" in response to any corporate-unfriendly inputs.
MiniGPT-4
261–270 of 337 posts
Re: MiniGPT-4
#262On a technical level, they're doing something really simple -- take BLIP2's ViT-L+Q-former, connect it to Vicuna-13B with a linear layer, and train just the tiny layer on some datasets of image-text pairs. But the results are pretty amazing. It completely knocks Openflamingo && even the original blip2 models out of the park. And best of all, it arrived before OpenAI's GPT-4 Image Modality did. Real win for Open Sourc…
>so hopefully by tomorrow it'll be runnable by 3090/4090 users. Taking a step back, this is just a wild statement. I know there's some doom and gloom out there, but in certain aspects, it's an awesome time to be alive.
It already runs!
Re: MiniGPT-4
#263It doesn't seem to recognize text from a screenshot very well, I gave detailed error messages from a Windows screenshot: https://filestore.community.support.microsoft.com/api/images... and prompted "Describe any issues found in this screenshot and steps to resolve them" while it correctly identified it as a screenshot from a computer, it gave a very generic response and didn't identify the error messages correctly: "…
I can barely read the text in that screenshot, and I'm a human (I think).
It is both low resolution and filled with jpeg artifacts.
Re: MiniGPT-4
#264Looking forward to the next generation of cheap GPUs with enough VRAM to run models like Vicuna-13 locally.
Vicuna-13B in GPTQ 4bit has almost no perplexity/quality loss and fits in just 8GB of RAM or VRAM.
I run it on my phone CPU and get ~4 tokens per second. On my laptop CPU I get 8 tokens per second.
On a $200 P40 I run LLaMA-33B at 12 tokens per second in GPTQ 4bit. A consumer 3090 gets over 20 tokens per second for LLaMA-33B and 30 tokens/second for Vicuna-13B.
Re: MiniGPT-4
#265Looking forward to the next generation of cheap GPUs with enough VRAM to run models like Vicuna-13 locally.
Cheap is relative I suppose. I’m running Vicuna 13b 16f locally and it needs 26GB of VRAM, which won’t even fit on a single RTX 4090. The next gen RTX Titan might have enough vram but that won’t come cheap. I’m expecting a price point above $2500.
13B in GPTQ 4bit has practically no quality loss and runs in 8GB. I get 8 tokens/second on my laptop CPU and 4 tokens/second on my phone CPU.
Even 33B only needs 20GB of VRAM in GPTQ 4bit.
8bit has zero perplexity loss, so there's really no reason to run in 16bit.
Even a $200 P40 24GB is enough to run 33B at extremely high speeds in GPTQ 4bit.
Re: MiniGPT-4
#266Earlier quoted context omitted.
I don't see how it's misleading. MiniGPT-4 makes it sound like a smaller alternative to GPT-4, if it was based on GPT-4 there would be nothing 'mini' about it.
It has more in common with GPT-3 than GPT-4 in terms of size, but in reality it's based on Vicuna/Llama which is 10x smaller than either, so as far as the LLM part of it goes its not mini-anything - it's just straight-up Vicuna 13B. The model as a whole is just BLIP-2 with a larger linear layer, and using Vicuna as the LLM. If you look at their code it's literally using the entire BLIP-2 encoder (Salesforce code). ht…
vicuna was done with sharegpt transcripts, right? did they ever say if those transcripts were while users were using gpt3.5 or gpt4.0?
Re: MiniGPT-4
#267Earlier quoted context omitted.
Hey, guys. Hey. Ready to talk plate processing and residue transport plate funneling? Why don't we start with joust jambs? Hey, why not? Plates and jousts. Can we couple them? Hell, yeah, we can. Want to know how? Get this. Proprietary to McMillan. Only us. Ready? We fit Donnely nut spacing grip grids and splay-flexed brace columns against beam-fastened derrick husk nuts and girdle plate Jerries, while plate flex tan…
This post is double great and I will never forgive Amazon for canceling that show. For those that don't know this is from a show called Patriot. https://en.wikipedia.org/wiki/Patriot_(TV_series) Scene: https://youtube.com/watch?v=-F-IHvF5OCA
I remember seeing someone link to that scene recently as a joke on Twitter (about Twitter trying to explain Twitter Blue). Within a few days I’d watched the entire series… absolutely phenomenal show.
Edit: ah I actually saw the prior scene where Leslie was explaining to John what he expected (which is the setup for the linked bit): https://www.youtube.com/watch?v=G7Do2tlYLhs
Re: MiniGPT-4
#268Re: MiniGPT-4
#269Earlier quoted context omitted.
> they're doing something really simple -- take BLIP2's ViT-L+Q-former, connect it to Vicuna-13B with a linear layer, and train just the tiny layer on some datasets of image-text pairs Oh yes. Simple! Jesus, this ML stuff makes a humble web dev like myself feel like a dog trying to read Tolstoy.
FWIW I work in LLMs and I consistently fail to do simple webdev stuff
Maybe you're just holding it wrong: You're not supposed to let your LLM rest or chat idly while you do the webdev stuff yourself, but to make your LLM do the webdev stuff for you ;P
Re: MiniGPT-4
#270Earlier quoted context omitted.
Wow, that must be an expensive domain name.
I'm sure they can afford it But justai.com would also be apt
If the criterion is to go with what's the most apt fit for the company, then "closedai.com" would be even better, and as a bonus it would probably be even cheaper… (even if they have enough money for a more expensive domain)