ttt3ts

13 ms·

91.
by ttt3ts 3y ago
It isn't script kiddie level but it isn't hard I finetuned a 15B parameter reddit bot with an afternoon of time and a day of training on a 3090. Bot got a few thousand Karma in a couple of days before I turned it off (proof of con
92.
by ttt3ts 3y ago
Do you know a dataset for fine-tuning roleplaying?
93.
by ttt3ts 3y ago
https://huggingface.co/blog/stackllama You can easily finetune 7B or 15B LORA model with that on consumer GPUs.
94.
by ttt3ts 3y ago
Join a different instance with a different color scheme?
95.
by ttt3ts 3y ago
Also, that isn't exactly a powerful machine at 4 cores and 16GB of ram. Probably have to double it or what is the point? Security, maybe?
96.
by ttt3ts 3y ago
I own a P100, P4, M40. They either lack ram or speed. Also, unless you are putting them in a server you have to cool them. If your goal is to learn ML don't tinker with very obsolete hardware. Rent or buy something modern.
97.
by ttt3ts 3y ago
No clue but if you want learn/finetuned ML use a Linux box otherwise you will spend all your time fighting your machine. If you just want to run models Mac might work.
98.
by ttt3ts 3y ago
Source? I own 3 and I would take a single 3090 any day. M40s are simply too old.
99.
by ttt3ts 3y ago
You can finetune whisper, stable diffusion, and LLM up to about 15B parameters with 24GB VRAM. Which leads you to what hardware to get. Best bang for the $ right now is definitely a used 3090 at ~$700. If you want more than 24GB vram just r
100.
by ttt3ts 3y ago
Thanks for the clarification. Looks like you're a new y combinator startup. Kool. I would suggest you find away to allow existing code to be reused. Deno market share is very small and experienced/larger teams are not going to wan
101.
by ttt3ts 3y ago
Was hoping it allowed my runtime. I have a bunch of scripts in NodeJS, Python 2, C#, etc. I don't care to rewrite them and/or write scripts tied to a weird sandbox runtime. Any support for docker based workers?
102.
by ttt3ts 3y ago
Finetuning which can easily be done on consumer hardware and can give these models a lot more power for specific applications. Also, ChatGPT just can't do a lot of things because of their "rules". I was doing question answeri
103.
by ttt3ts 3y ago
Not a good comparison. CPU vs TPU. Also, whisper.cpp perf varies a ton depending on hardware. Avx512 or neural engine?
104.
by ttt3ts 3y ago
Yea, whisper runs fairly quickly on really old GPUs and leveraging CTranslate2 (underlying tech in your link) you need < 4GB of GPU RAM to run the large model. This might be faster but an order of magnitude more expensive? Also, huge loc
105.
by ttt3ts 4y ago
+1 for academic torrents. Torrents are a lot more mature and has a better culture around seeding to ensure models and data stay alive.
106.
by ttt3ts 4y ago
No, was scraping TBs over it when NFTs we're a thing. It murders CPU, garbage cleanup is so bad you should just disable it, causes network issues. Seems like it's momentum is gone and if you haven't interacted with it yet you
107.
by ttt3ts 4y ago
Nvidia's last gen...
108.
by ttt3ts 4y ago
Faiss supports FLAT aka not approx nearest neighbor. Also, all the vector databases use ANN too because nearest neighbor is hard. Or are you referring to the ability to store data in addition to the vectors? In which case, you can pair any
109.
by ttt3ts 4y ago
Scale, Faiss supports a bunch more algorithms. A much larger scale.
110.
by ttt3ts 4y ago
"Cheaping out" is not the best description. Apple has riddled the space with a ton of patents. https://insights.greyb.com/apple-facial-recognition-patents/
111.
by ttt3ts 4y ago
So you can? (serious question not sarcastic)
112.
by ttt3ts 4y ago
Or the world has always had issues and the lense through which we now view the world is an everpresent fear coaster. Step away from the news & social media for a few months. You can't effect the outcome anyway and you might find yo
113.
by ttt3ts 4y ago
Lol, not quite that jank yet. Maybe in another few years.
114.
by ttt3ts 4y ago
Don't worry the butt warmer will have failed long ago in your 15 year old EV Typed while sitting in my 20 year old pickup truck whos manual windows still work but not most of the electronics :P
115.
by ttt3ts 4y ago
Redis uses too much memory and supports only two NN algorithms FLAT (very slow) and HNSW if you start indexing millions of large vectors you will quickly understand the problem. That being said, many of these DBs are overcomplicated for mos
116.
by ttt3ts 4y ago
Don't start with mullivus if you're learning. Too much yak shaving. Try https://github.com/criteo/autofaiss . Also, TBH, it is a lot cheaper to run a simple faiss index.
117.
by ttt3ts 4y ago
Learn faiss first. Mullvus is very complicated in comparison and faiss will work in many cases. I mean look at their system diagram https://milvus.io/static/0bc2e74d0a1b20bbfb91bdbd03f77e5e/bb... . Not fun...
118.
by ttt3ts 4y ago
I don't care what Meta does they are big enough to solve the issue. I was talking about smaller messaging apps that can't just refuse to comply.
119.
by ttt3ts 4y ago
Just block the UK and make everyone use a VPN this is already the case with age verification for adult content.
120.
by ttt3ts 4y ago
Who cares? Expected lifetime CO2 for an F150 is 100 tons. So, equivalent to 10 more pickup trucks. Seems like the progress is worth the cost.

More ›