Before you get your hopes up: Anyone can download it, but very few will be able to actually run it.
What’s the resources requirements for the 405B model? I did some digging but couldn’t find any documentation during my cursory search.
Typically you need about 1GB graphics RAM for each billion parameters (i.e. one byte per parameter). This is a 405B parameter model. Ouch.
Edit: you can try quantizing it. This reduces the amount of memory required per parameter to 4 bits, 2 bits or even 1 bit. As you reduce the size, the performance of the model can suffer. So in the extreme case you might be able to run this in under 64GB of graphics RAM.
Typically you need about 1GB graphics RAM for each billion parameters (i.e. one byte per parameter). This is a 405B parameter model.
Or you could run it via cpu and ram at a much slower rate.
Yeah uh let me just put in my 512GB ram stick…
Samsung do make them.
Goodluck finding 512gb of VRAM.
https://www.ebay.com/p/116332559 lga2011 motherboards quite cheap, insert 2 xeon 2696v4 44 threads each totalling at 88 threads and 8 ddr4 32gb sticks, it comes quite cheap actually, you can also install Nvidia p40 with 24gb each, you can max out this build for ai for under 2000$
Finally! My dumb dumb 1TB ram server (4x E5-4640 + 32x32GB DDR3 ECC) can shine.
At work we habe a small cluster totalling around 4TB of RAM
It has 4 cooling units, a m3 of PSUs and it must take something like 30 m2 of space
When the 8 bit quants hit, you could probably lease a 128GB system on runpod.
Can you run this in a distributed manner, like with kubernetes and lots of smaller machines?
According to huggingface, you can run a 34B model using 22.4GBs of RAM max. That's a RTX 3090 Ti.
Ypu mean my 4090 isn't good enough 🤣😂
Hmm, I probably have that much distributed across my network... maybe I should look into some way of distributing it across multiple gpu.
Frak, just counted and I only have 270gb installed. Approx 40gb more if I install some of the deprecated cards in any spare pcie slots i can find.
405b ain't running local unless you got a proepr set up is enterpise grade lol
I think 70b is possible but I haven't find anyone confirming it yet
Also would like to know specs on whoever did it
I've run quantized 70B models on CPU with 32 gigs but it is very slow
I gonna add some RAM with hope I can split original 70b between GPU and RAM. 8b is great what it is as is
Looks like it should be possible, not sure how much performance hit offloading to RAM will do. Fafo
I have a home server with 140 gigs of RAM, it was surprisingly cheap. It's an HP z6 with the 6146 gold xeon processor.
I found a seller who was selling it with a low spec silver and 16 gigs of RAM for like 250 bucks.
Found the processor upgrade for about $120 and spend another $150 on 128gb of second-hand ECC ddr4.
I think the total cost was something like $700 after throwing a couple of 8 TB hard drives in.
I've also placed a Nvidia 4070 in it, which I got doing some horse trading.
How close am I on the specs to being able to run the 70b version?
What's the bus speed of the RAM? You might run it just fine but still bottlenecked there.
It's clocked at ddr4 2666
With 144Gb of total RAM, you should be able to run any CPU intensive software.
The LLMs use GPU vRAM though, so it doesn't matter how much system RAM you have, since GPU vRAM is what the xformers and tensor scripts prioritize and have been ultimately optimized to use over CPU and RAM.
I regularly run llama3 70b unqantized on two P40s and CPU at like 7tokens/s. It's usable but not very fast.
ok this is a server. 48gb cards and 67gb ram? for model alone?
Each card has 24GB so 48GB vram total. I use ollama it fills whatever vrams is available on both cards and runs the rest on the CPU cores.
What are you asking exactly?
What do you want to run? I assume you have a 24GB GPU and 64GB host RAM?
correct. and how ram speed work in this tbh
My memory sticks are all DDR4 with 32GB@2133MT/s.
As a general rule of thumb, you need about 1 GB per 1B parameters, so you're looking at about 405 GB for the full size of the model.
Quantization can compress it down to 1/2 or 1/4 that, but "makes it stupider" as a result.
This would probably run on a a6000 right?
Edit: nope I think I'm off by an order of magnitude
"an order of magnitude" still feels like an understatement LOL
My 35b models come out at like Morse code speed on my 7800XT, but at least it does work?
When the RTX 9090 Ti comes, anyone who can afford it will be able to run it.
That doesn't sound like much of a change from the situation right now.
So does OSM data. Everyone can download the whole earth but to serve it and provide routing/path planning at scale takes a whole other skill and resources. It's a good thing that they are willing to open source their model in the first place.
Wake me up when it works offline
"The Llama 3.1 models are available for download through Meta's own website and on Hugging Face. They both require providing contact information and agreeing to a license and an acceptable use policy, which means that Meta can technically legally pull the rug out from under your use of Llama 3.1 or its outputs at any time."
WAKE UP!
It works offline. When you use with ollama, you don't have to register or agree to anything.
Once you have downloaded it, it will keep on working, meta can't shut it down.
Well, yes and no. See the other comment, 64 GB VRAM at the lowest setting.
Oh, sure. For the 405B model it's absolutely infeasible to host it yourself. But for the smaller models (70B and 8B), it can work.
I was mostly replying to the part where they claimed meta can take it away from you at any point - which is simply not true.
It's available through ollama already. i am running the 8b model on my little server with it's 3070 as of right now.
It's really impressive for a 8b model
Intriguing. Is that an 8gb card? Might have to try this after all
Yup, 8GB card
Its my old one from the gaming PC after switching to AMD.
It now serves as my little AI hub and whisper server for home assistant
What the heck is whisper? Ive been fooling around with hass for ages, haven't heard of it even after at least two minutes of searching. Is it openai affiliated hardwae?
whisper is an STT application that stems from openAI afaik, but it's open source at this point.
It did ask to make an account but I didn't see any pinging home when I did it.
What am I missing here?
Through meta...
That's where I stop caring
Yo this is big. In both that it is momentous, and holy shit that’s a lot of parameters. How many GB is this model?? I’d be able to run it if I had an few extra $10k bills lying around to buy the required hardware.
its around 800gb
God damn.
That's some thick model
Time to buy a thread ripper and 800gb of ram so that I can run this model at 1 token per hour.
Kind of petty from Zuck not to roll it out in Europe due to the digital services act.. But also kind of weird since it's open source? What's stopping anyone from downloading the model and creating a web ui for Europe users?
Did anyone get 70b to run locally?
If so what, what hardware specs?
Afaik you need about 40GB of vram for a 70b model.
Can't you offload some of it to RAM?
Same requirements, but much slower.
I guess time to buy some ram after spending decade at 16gb
That looks good on paper, but while I find ChatGPT good to create critical thinking, I've found Meta's products (Facebook and Instagram) to be sources of disinformation. That makes me have reservations about Meta's intentions with LLMs. As the article says, the model comes pre-trained, so it's most made up of information gathered by Meta.
Neither Meta nor anyone else is hand-curating their dataset. The fact that Facebook is full of grandparents sharing disinformation doesn't impact what's in their model.
But all LLMs are going to have accuracy issues because they're 1) trained on text written by humans who themselves are inaccurate and 2) designed to choose tokens based on probability rather than any internal logic as to whether an answer is factual.
All LLMs are full of shit. That doesn't mean they're not fun or even useful in some applications, but you shouldn't trust anything they write.
Technically correct (tm)
Before you get your hopes up: Anyone can download it, but very few will be able to actually run it.
What’s the resources requirements for the 405B model? I did some digging but couldn’t find any documentation during my cursory search.
Typically you need about 1GB graphics RAM for each billion parameters (i.e. one byte per parameter). This is a 405B parameter model. Ouch.
Edit: you can try quantizing it. This reduces the amount of memory required per parameter to 4 bits, 2 bits or even 1 bit. As you reduce the size, the performance of the model can suffer. So in the extreme case you might be able to run this in under 64GB of graphics RAM.
Or you could run it via cpu and ram at a much slower rate.
Yeah uh let me just put in my 512GB ram stick…
Samsung do make them.
Goodluck finding 512gb of VRAM.
https://www.ebay.com/p/116332559 lga2011 motherboards quite cheap, insert 2 xeon 2696v4 44 threads each totalling at 88 threads and 8 ddr4 32gb sticks, it comes quite cheap actually, you can also install Nvidia p40 with 24gb each, you can max out this build for ai for under 2000$
Finally! My dumb dumb 1TB ram server (4x E5-4640 + 32x32GB DDR3 ECC) can shine.
At work we habe a small cluster totalling around 4TB of RAM
It has 4 cooling units, a m3 of PSUs and it must take something like 30 m2 of space
When the 8 bit quants hit, you could probably lease a 128GB system on runpod.
Can you run this in a distributed manner, like with kubernetes and lots of smaller machines?
According to huggingface, you can run a 34B model using 22.4GBs of RAM max. That's a RTX 3090 Ti.
Ypu mean my 4090 isn't good enough 🤣😂
Hmm, I probably have that much distributed across my network... maybe I should look into some way of distributing it across multiple gpu.
Frak, just counted and I only have 270gb installed. Approx 40gb more if I install some of the deprecated cards in any spare pcie slots i can find.
405b ain't running local unless you got a proepr set up is enterpise grade lol
I think 70b is possible but I haven't find anyone confirming it yet
Also would like to know specs on whoever did it
I've run quantized 70B models on CPU with 32 gigs but it is very slow
I gonna add some RAM with hope I can split original 70b between GPU and RAM. 8b is great what it is as is
Looks like it should be possible, not sure how much performance hit offloading to RAM will do. Fafo
I have a home server with 140 gigs of RAM, it was surprisingly cheap. It's an HP z6 with the 6146 gold xeon processor.
I found a seller who was selling it with a low spec silver and 16 gigs of RAM for like 250 bucks.
Found the processor upgrade for about $120 and spend another $150 on 128gb of second-hand ECC ddr4.
I think the total cost was something like $700 after throwing a couple of 8 TB hard drives in.
I've also placed a Nvidia 4070 in it, which I got doing some horse trading.
How close am I on the specs to being able to run the 70b version?
What's the bus speed of the RAM? You might run it just fine but still bottlenecked there.
It's clocked at ddr4 2666
With 144Gb of total RAM, you should be able to run any CPU intensive software.
The LLMs use GPU vRAM though, so it doesn't matter how much system RAM you have, since GPU vRAM is what the xformers and tensor scripts prioritize and have been ultimately optimized to use over CPU and RAM.
I regularly run llama3 70b unqantized on two P40s and CPU at like 7tokens/s. It's usable but not very fast.
so there is no way a 24gb and 64gb can run thing?
My specs because you asked:
ok this is a server. 48gb cards and 67gb ram? for model alone?
Each card has 24GB so 48GB vram total. I use ollama it fills whatever vrams is available on both cards and runs the rest on the CPU cores.
What are you asking exactly?
What do you want to run? I assume you have a 24GB GPU and 64GB host RAM?
correct. and how ram speed work in this tbh
My memory sticks are all DDR4 with 32GB@2133MT/s.
As a general rule of thumb, you need about 1 GB per 1B parameters, so you're looking at about 405 GB for the full size of the model.
Quantization can compress it down to 1/2 or 1/4 that, but "makes it stupider" as a result.
This would probably run on a a6000 right?
Edit: nope I think I'm off by an order of magnitude
"an order of magnitude" still feels like an understatement LOL
My 35b models come out at like Morse code speed on my 7800XT, but at least it does work?
When the RTX 9090 Ti comes, anyone who can afford it will be able to run it.
That doesn't sound like much of a change from the situation right now.
So does OSM data. Everyone can download the whole earth but to serve it and provide routing/path planning at scale takes a whole other skill and resources. It's a good thing that they are willing to open source their model in the first place.
Wake me up when it works offline "The Llama 3.1 models are available for download through Meta's own website and on Hugging Face. They both require providing contact information and agreeing to a license and an acceptable use policy, which means that Meta can technically legally pull the rug out from under your use of Llama 3.1 or its outputs at any time."
WAKE UP!
It works offline. When you use with ollama, you don't have to register or agree to anything.
Once you have downloaded it, it will keep on working, meta can't shut it down.
Well, yes and no. See the other comment, 64 GB VRAM at the lowest setting.
Oh, sure. For the 405B model it's absolutely infeasible to host it yourself. But for the smaller models (70B and 8B), it can work.
I was mostly replying to the part where they claimed meta can take it away from you at any point - which is simply not true.
It's available through ollama already. i am running the 8b model on my little server with it's 3070 as of right now.
It's really impressive for a 8b model
Intriguing. Is that an 8gb card? Might have to try this after all
Yup, 8GB card
Its my old one from the gaming PC after switching to AMD.
It now serves as my little AI hub and whisper server for home assistant
What the heck is whisper? Ive been fooling around with hass for ages, haven't heard of it even after at least two minutes of searching. Is it openai affiliated hardwae?
whisper is an STT application that stems from openAI afaik, but it's open source at this point.
i wrote a little guide on how to install it on a server with an NVidia GPU and hw acceleration and integrate it into your homeassistant after. https://a.lemmy.dbzer0.com/lemmy.dbzer0.com/comment/5330316
it's super fast with a GPU available and i use those little M5 ATOM Echo microphones for this.
I'm running 3.1 8b as we speak via ollama totally offline and gave info to nobody.
https://ollama.com/library/llama3.1
I was able to set up small one via open webui.
It did ask to make an account but I didn't see any pinging home when I did it.
What am I missing here?
Through meta...
That's where I stop caring
Yo this is big. In both that it is momentous, and holy shit that’s a lot of parameters. How many GB is this model?? I’d be able to run it if I had an few extra $10k bills lying around to buy the required hardware.
its around 800gb
God damn.
That's some thick model
Time to buy a thread ripper and 800gb of ram so that I can run this model at 1 token per hour.
Kind of petty from Zuck not to roll it out in Europe due to the digital services act.. But also kind of weird since it's open source? What's stopping anyone from downloading the model and creating a web ui for Europe users?
Did anyone get 70b to run locally?
If so what, what hardware specs?
Afaik you need about 40GB of vram for a 70b model.
Can't you offload some of it to RAM?
Same requirements, but much slower.
I guess time to buy some ram after spending decade at 16gb
That looks good on paper, but while I find ChatGPT good to create critical thinking, I've found Meta's products (Facebook and Instagram) to be sources of disinformation. That makes me have reservations about Meta's intentions with LLMs. As the article says, the model comes pre-trained, so it's most made up of information gathered by Meta.
Neither Meta nor anyone else is hand-curating their dataset. The fact that Facebook is full of grandparents sharing disinformation doesn't impact what's in their model.
But all LLMs are going to have accuracy issues because they're 1) trained on text written by humans who themselves are inaccurate and 2) designed to choose tokens based on probability rather than any internal logic as to whether an answer is factual.
All LLMs are full of shit. That doesn't mean they're not fun or even useful in some applications, but you shouldn't trust anything they write.