@aistartupfren: How much money did training Llama 3.1 cost? #aistartup #techstartup #llama #llm #ai

Lan 🌊 AI startup friend
Lan 🌊 AI startup friend
Open In TikTok:
Region: US
Friday 06 September 2024 16:20:00 GMT
19064
505
37
56

Music

Download

Comments

dadjokesandrandomn
dadjokesandrandomness :
big tech companies don't pay retail rates for stuff. they bulk buy electricity and GPUs, probably 40-60% off public pricing
2024-09-06 18:48:45
19
onlypooches
OnlyDogs :
They get deals on power cause they use so much, and bulk purchasing power on chips. Don’t forget the tax incentives for the data center as well..
2024-09-07 00:37:19
5
drpillows
Pillows :
who is using a 700 watt microwave 😆😆😆😆
2024-09-07 01:21:53
4
brannisq360
BranNisq360 :
That’s not including those industrial fans for cooling and power generation
2024-09-06 22:11:40
3
rayylad
Rayy Lad :
I’m sure it was trained on H100s and those are way more expensive to run and buy/rent
2024-09-07 15:53:17
1
tiktoker1000101101
tiktoker1001 :
5 million is actually cheap, the engineering can be more expensive
2024-09-06 19:06:07
2
yungwill_
will :
Ai is gonna make us have to come up with better power solutions
2024-09-06 17:05:43
2
chicago_summerlover
Chicago_summer_fan :
Dont worry about people defining watts and all. We just need a high level cost which is covered. Thank for making the video. Good content
2024-09-28 06:47:03
2
chadmoran
Chad Moran :
Watt is a rate not a measurement of capacity.
2024-09-08 07:05:43
1
lanefowler
Lane Fowler :
40 million micro wave ovens for 1 hour, that's a lot of popcorn!
2024-09-07 14:20:19
1
5gjay
parasoxical :
doesn't this appear suspiciously like Bitcoin mining
2024-09-07 23:20:14
1
alittlepositivityyy
ALittlePositivity :
I love how u break that down
2024-09-30 05:19:29
1
cgt3103
cgt3103 :
That’s an entire power plants output. Just training one model.
2024-10-04 18:28:02
1
vicksemmanuel1
vicksemmanuel1 :
Detailed breakdown
2024-09-06 16:42:21
2
snoww131
John :
can you do a video just doing a explainer on GPU hours and history of how we came to the term?
2024-09-09 18:14:50
2
aiwebtools
AiWebTools :
💯very informative thanks
2024-09-06 22:18:21
2
forabetterworld24
Eigenclass ✊ 🔻 :
You’re confusing watt, a unit of power and watt-hour, a unit of energy. The latter is what you want, the former doesn’t have a price
2024-09-07 08:05:00
1
lanefowler
Lane Fowler :
Fantastic analogy! I'm going to share this with some others. I'm following you now
2024-09-07 14:19:14
1
freek_im_under_your_bed
freek_im_under_your_bed :
They're training on Azure servers, which is owned by Microsoft, and that cost is covered by Microsoft. They're giving them billion dollars worth of compute time for special access
2024-09-07 01:57:33
1
9jadev
Don dada :
A lottttt 😁
2024-11-28 13:04:09
0
deespotatos
b_random :
So, are we gonna start stretching our training farms out to Antarctica soon? Or does the cost of remote power plants outweigh the cost of free cooling?
2024-09-07 16:17:05
0
u_are_a_turd
YouAreATurd :
I train ML models at scale (usually not LLMs but GBMs) and even at a small bank tuning/training a valuations model costs thousands of dollars.
2024-09-07 01:04:00
0
oldewhittington
Whittington III™ 🤗 :
They have their own data centers so it’s possible they used their own hardware 🤔
2024-09-06 20:53:10
0
by123808
by123808 :
Hold up! Which website rent A100 for only $1/hr?
2024-09-06 19:11:59
0
ninjaducksauce
NinjaDuckSauce :
My llamas eat grass and training is free
2024-09-07 00:08:08
2
To see more videos from user @aistartupfren, please go to the Tikwm homepage.

Other Videos


About