Загрузка видео...

Не удалось загрузить видео

На главную

llm.c by Hand✍️ C programming + matrix multiplication by hand This combination is perhaps as low as we can get to explain how the Transformer works. Special thanks to Andrej Karpathy for encouraging early feedback and tetsuo //: 👾 for helping me understand the pragma magic. I hope this...

ProfTomYeh's profile picture

Tom Yeh

65,480 subscribers

311,792 просмотров • 2 лет назад •via X (Twitter)

ОбразованиеНовости и политикаНаука и технологии
LIVE

Anya Rossi• Live Now

Private livecam show

Комментарии: 11

Фото профиля Yuchen Jin
Yuchen Jin2 лет назад

@karpathy @7etsuo Awesome! Looking forward to an llm.c Attention episode

Фото профиля Tom Yeh
Tom Yeh2 лет назад

@karpathy @7etsuo This is a great suggestion! Thanks!

Фото профиля Antonio Linares
Antonio Linares2 лет назад

@karpathy @7etsuo Is there a way to load a single layer of a LLM (in example Llama3-70B) and process it layer by layer so it fits on a normal PC memory ? Could you explain how to do it ?

Фото профиля Tom Yeh
Tom Yeh2 лет назад

@karpathy @7etsuo It is possible. But it would be quite slow. To generate each token, the model has to evaluate every one of the hundreds of layers. Say loading each layer takes 1 second. It would take 100 seconds to load 100 layers for one token. Then you multiple that by 1000 tokens. Very slow.

Фото профиля rakesh
rakesh2 лет назад

@karpathy @7etsuo Goated acc

Фото профиля ed
ed2 лет назад

@karpathy @7etsuo Amazing 🔥

Фото профиля Cuhmunity Notes
Cuhmunity Notes2 лет назад

@karpathy @7etsuo 🔥

Фото профиля Byron Hsu
Byron Hsu2 лет назад

@karpathy @7etsuo This is awesome

Фото профиля Filippo Broggini
Filippo Broggini2 лет назад

@karpathy @7etsuo Is this also available elsewhere online?

Фото профиля George D Gregory
George D Gregory2 лет назад

@karpathy @7etsuo Can these be made into the notebooks!? That would be awesome.

Фото профиля Tom Yeh
Tom Yeh2 лет назад

@karpathy @7etsuo I am waiting for a notebook ninja to step up to help.😄

Похожие видео

[Transformer] by Hand✍️📺 5-minute Video Tutorial Anna Rahn made this short video to explain the Transformer exercise for my Computer Vision course last spring. In 5 minutes, she demonstrates the key calculations of the Transformer by hand with pen and paper! Anna is a fantastic student. I am lucky to have her in my lab!
4:55

[Transformer] by Hand✍️📺 5-minute Video Tutorial Anna Rahn made this short video to explain the Transformer exercise for my Computer Vision course last spring. In 5 minutes, she demonstrates the key calculations of the Transformer by hand with pen and paper! Anna is a fantastic student. I am lucky to have her in my lab!

Tom Yeh

133,693 просмотров • 2 лет назад

Can you follow the math of a Transformer model by hand ✍️? Download PDF 👉 I hope this hand-drawn animation makes it easier. p.s., This has one error. Can you can spot it and tell me in the comment?
0:30

Can you follow the math of a Transformer model by hand ✍️? Download PDF 👉 I hope this hand-drawn animation makes it easier. p.s., This has one error. Can you can spot it and tell me in the comment?

Tom Yeh

14,578 просмотров • 1 год назад

Self-Attention by hand ✍️ Excel ~ I designed this exercise for students to practice the QKV math. I also created a medium and a large version to show how the attention matrix grows quadratically as the sequence gets longer. 👇Join the 'AI Math' community. Download xlsx.
0:20

Self-Attention by hand ✍️ Excel ~ I designed this exercise for students to practice the QKV math. I also created a medium and a large version to show how the attention matrix grows quadratically as the sequence gets longer. 👇Join the 'AI Math' community. Download xlsx.

Tom Yeh

125,723 просмотров • 2 лет назад

Anna Rahn's video demonstration of Multi-Layer Perceptron (MLP) by hand ✍️ Anna made this short video for my computer vision course in the spring semester. Together with the spreadsheet exercise I posted yesterday, I hope they help more people understand deep neural networks!
3:35

Anna Rahn's video demonstration of Multi-Layer Perceptron (MLP) by hand ✍️ Anna made this short video for my computer vision course in the spring semester. Together with the spreadsheet exercise I posted yesterday, I hope they help more people understand deep neural networks!

Tom Yeh

120,565 просмотров • 2 лет назад

Mamba by hand ✍️ ~ I made this exercise as I studied the math in the Mamba paper. It is a small yet complete Mamba implementation. Thanks Albert Gu for answering my questions and checking. 👇Join the 'AI Math' community. Download xlsx.
0:21

Mamba by hand ✍️ ~ I made this exercise as I studied the math in the Mamba paper. It is a small yet complete Mamba implementation. Thanks Albert Gu for answering my questions and checking. 👇Join the 'AI Math' community. Download xlsx.

Tom Yeh

55,484 просмотров • 2 лет назад

We all smoke our way, for now I have mastered the art of the glass tip cone style. It’s not a specialty roll by any means, but it’s consistent for me. I don’t have the fine hand coordination nor the patience to fuck up a hand roll when I can do this pretty easily and it smokes as proper as a good hand roll. When I have more time away from serving, I’ll practice hand rolling but this is how I smoke.
0:18

Sensitive content

We all smoke our way, for now I have mastered the art of the glass tip cone style. It’s not a specialty roll by any means, but it’s consistent for me. I don’t have the fine hand coordination nor the patience to fuck up a hand roll when I can do this pretty easily and it smokes as proper as a good hand roll. When I have more time away from serving, I’ll practice hand rolling but this is how I smoke.

Don

22,510 просмотров • 8 месяцев назад

Can you calculate a Vision Transformer (ViT) by hand? ✍️ If you can, come join me next week at the LLM Paper Club. I will extend ViT to Llama 1 -> 2 -> 3 -> 4 live. Event link below.
0:34

Can you calculate a Vision Transformer (ViT) by hand? ✍️ If you can, come join me next week at the LLM Paper Club. I will extend ViT to Llama 1 -> 2 -> 3 -> 4 live. Event link below.

Tom Yeh

79,441 просмотров • 1 год назад

It’s hard to post this cruelty, this hate, this evil, as it makes me so angry - but it needs to be seen. We need to show what Trump is doing to our people. We are fighting Trump and ICE by getting into the hands of as many immigrants and allies as possible. As well as lawful and constitutional immigration advice, PAXIS can help people prepare for conflict, and can help defuse situations. FIGHT ICE, FIGHT TRUMP, FIGHT FASCISM:
1:51

It’s hard to post this cruelty, this hate, this evil, as it makes me so angry - but it needs to be seen. We need to show what Trump is doing to our people. We are fighting Trump and ICE by getting into the hands of as many immigrants and allies as possible. As well as lawful and constitutional immigration advice, PAXIS can help people prepare for conflict, and can help defuse situations. FIGHT ICE, FIGHT TRUMP, FIGHT FASCISM:

Joe Walsh

19,776 просмотров • 7 месяцев назад

The London promise is simple. If you work hard, and get a helping hand, you can achieve anything. That's why I'm running again as Mayor: so City Hall can continue to be that helping hand for every Londoner.
0:34

The London promise is simple. If you work hard, and get a helping hand, you can achieve anything. That's why I'm running again as Mayor: so City Hall can continue to be that helping hand for every Londoner.

Sadiq Khan

198,631 просмотров • 2 лет назад

"I've been told by multiple sources that we can expect [Joe Biden] to make his announcement as early as the weekend, perhaps next week, and to clear the way for Kamala Harris and perhaps others to compete for the nomination." — Mark Halperin, political analyst Rob Finnerty
5:00

"I've been told by multiple sources that we can expect [Joe Biden] to make his announcement as early as the weekend, perhaps next week, and to clear the way for Kamala Harris and perhaps others to compete for the nomination." — Mark Halperin, political analyst Rob Finnerty

NEWSMAX

65,255 просмотров • 2 лет назад

1/ Mortgage rates in Mozambique have reached as high as 22%. Much of the population along the coastline can’t get a loan to buy a house, and shelters built by hand get destroyed by cyclones. A quick look at how crypto can update the system to help get people in safer homes.🧵
1:00

1/ Mortgage rates in Mozambique have reached as high as 22%. Much of the population along the coastline can’t get a loan to buy a house, and shelters built by hand get destroyed by cyclones. A quick look at how crypto can update the system to help get people in safer homes.🧵

Coinbase 🛡️

613,572 просмотров • 3 лет назад

i simply don't understand why nobody is building this... Andrej Karpathy, OpenAI Co-founder, introduced the "LLM Wiki" - a knowledge graph inside Obsidian i built my own version for AI UGC and here's exactly how you can do the same: - start by scraping hundreds of ads to extract hooks, scripts, voiceovers and CTAs - prompt Claude Code to turn this into an Obsidian Vault - write scripts using Fable 5 and send these to Higgsfield Supercomputer - generate avatars using Nano Banana Pro and clips with Seedance the article below teaches you how to build the full workflow step-by-step:
0:26

i simply don't understand why nobody is building this... Andrej Karpathy, OpenAI Co-founder, introduced the "LLM Wiki" - a knowledge graph inside Obsidian i built my own version for AI UGC and here's exactly how you can do the same: - start by scraping hundreds of ads to extract hooks, scripts, voiceovers and CTAs - prompt Claude Code to turn this into an Obsidian Vault - write scripts using Fable 5 and send these to Higgsfield Supercomputer - generate avatars using Nano Banana Pro and clips with Seedance the article below teaches you how to build the full workflow step-by-step:

Machina

20,427 просмотров • 3 месяцев назад

CUDA MODE hackathon today! Here's Andrej Karpathy on the 🏖️ origin story of llm.c, and what it hints at for the fast, simple, llm-compiled future of custom software.
23:28

CUDA MODE hackathon today! Here's Andrej Karpathy on the 🏖️ origin story of llm.c, and what it hints at for the fast, simple, llm-compiled future of custom software.

swyx

97,516 просмотров • 2 лет назад

It’s time to understand why 63% enrichment of uranium is the magic number. I’ll let Maryam Shani explain as only she can. Don’t know about you but I am so much wiser for her bite size contributions to this important conversation as the U.S. looks to further talks in Islamabad. 🎥 TikTok -
2:59

It’s time to understand why 63% enrichment of uranium is the magic number. I’ll let Maryam Shani explain as only she can. Don’t know about you but I am so much wiser for her bite size contributions to this important conversation as the U.S. looks to further talks in Islamabad. 🎥 TikTok -

𝔗𝔯𝔲𝔱𝔥 𝔐𝔞𝔱𝔱𝔢𝔯𝔰

10,488 просмотров • 6 месяцев назад

What is the future of democracy worldwide in the era of #digitalization?🎧 Tune in as we delve into the pressing issue of how we can use digital tools to advocate for #peace and #democracy: ✨A special thanks to Nordic Innovation House - Tokyo for orchestrating this!
0:42

What is the future of democracy worldwide in the era of #digitalization?🎧 Tune in as we delve into the pressing issue of how we can use digital tools to advocate for #peace and #democracy: ✨A special thanks to Nordic Innovation House - Tokyo for orchestrating this!

The Nordics

25,520 просмотров • 2 лет назад

MLP by hand✍️Excel ~ I designed this exercise to show how to calculate a four-level network by hand, along with a graph and matching pytorch code. Also, I made a medium and a large version to show how it scales. 👇Join the 'AI Math' community. Download xlsx.
0:21

MLP by hand✍️Excel ~ I designed this exercise to show how to calculate a four-level network by hand, along with a graph and matching pytorch code. Also, I made a medium and a large version to show how it scales. 👇Join the 'AI Math' community. Download xlsx.

Tom Yeh

62,276 просмотров • 2 лет назад