A Hackers' Guide to Language Models
Vložit
- čas přidán 4. 07. 2024
- In this deeply informative video, Jeremy Howard, co-founder of fast.ai and creator of the ULMFiT approach on which all modern language models (LMs) are based, takes you on a comprehensive journey through the fascinating landscape of LMs. Starting with the foundational concepts, Jeremy introduces the architecture and mechanics that make these AI systems tick. He then delves into critical evaluations of GPT-4, illuminates practical uses of language models in code writing and data analysis, and offers hands-on tips for working with the OpenAI API. The video also provides expert guidance on technical topics such as fine-tuning, decoding tokens, and running private instances of GPT models.
As we move further into the intricacies, Jeremy unpacks advanced strategies for model testing and optimization, utilizing tools like GPTQ and Hugging Face Transformers. He also explores the potential of specialized datasets like Orca and Platypus for fine-tuning and discusses cutting-edge trends in Retrieval Augmented Generation and information retrieval. Whether you're new to the field or an established professional, this presentation offers a wealth of insights to help you navigate the ever-evolving world of language models.
(The above summary was, of course, created by an LLM!)
For the notebook used in this talk, see github.com/fastai/lm-hackers.
00:00:00 Introduction & Basic Ideas of Language Models
00:18:05 Limitations & Capabilities of GPT-4
00:31:28 AI Applications in Code Writing, Data Analysis & OCR
00:38:50 Practical Tips on Using OpenAI API
00:46:36 Creating a Code Interpreter with Function Calling
00:51:57 Using Local Language Models & GPU Options
00:59:33 Fine-Tuning Models & Decoding Tokens
01:05:37 Testing & Optimizing Models
01:10:32 Retrieval Augmented Generation
01:20:08 Fine-Tuning Models
01:26:00 Running Models on Macs
01:27:42 Llama.cpp & Its Cross-Platform Abilities
This is an extended version of the keynote given at posit::conf(2023). Thanks to @wolpumba4099 for chapter titles.
Gotta admit I'm feeling kinda teary reading all the lovely comments here. Thank you everybody -- love you all!
You just deserved it.😃
Second in the replies. :3
You are beyond awesome, Jeremy
Thanks for your work and please help us keep an eye on the apostles of the emerging noosphere, like Ben Goertzel ect.
Jeremy Thank you! This has helped so much. I’ve been a FastAI builder since the early days in 2017. Youre my hero. Appreciate all of the work you’ve done in the field
Just realised Jeremys paper led to the LLM revolution. Such a humble kind man. God bless you and all your students. You are such an example to follow. An example in character, humility and intelligence.
how ? which paper?
@@circleAI ULMFiT
@@circleAI Part of the answer is in the video's description.
Yes exactly what I was thinking, why there's so many people bragging about what they are doing and look at this guy, just helping others out.
This is probably the best invested youtube time of this year so far. What a gem. A lot of things he mentions have taken me month to figure out by my own. My new GPT-4 prompts will begin with "You are the expert Jeremy Howard..."
This!
lol, nice
100%
So great.
Absolutely!
*Transcript Summary:*
- Introduction & Basic Ideas of Language Models (00:00:00 - 00:18:05)
- Limitations & Improvements of GPT-4 (00:18:05 - 00:31:28)
- AI Applications in Code Writing, Data Analysis & OCR (00:31:28 - 00:38:50)
- Practical Tips on Using OpenAI API (00:38:50 - 00:46:36)
- Creating a Code Interpreter with Function Calling (00:46:36 - 00:51:57)
- Using Local Language Models & GPU Options (00:51:57 - 00:59:33)
- Fine-Tuning Models & Decoding Tokens (00:59:33 - 01:05:37)
- Testing & Optimizing Models with GPTQ & Hugging Face (01:05:37 - 01:09:48)
- Fine-Tuning with Llama 2 & Platypus Datasets (01:09:48 - 01:10:32)
- Retrieval Augmented Generation & Information Retrieval (01:10:32 - 01:20:08)
- Running a Private GPT & Fine-Tuning Models (01:20:08 - 01:22:32)
- Running Models on Macs (01:26:00 - 01:27:42)
- Discussing Llama.cpp & Its Cross-Platform Abilities (01:27:42 - 01:30:07)
- Challenges & Opportunities in Language Models (01:30:07 - 01:31:05)
Key points of interest: Function usage in GPT-4 (00:46:36), OCR application with Google Bard (00:33:59), and improving GPT-4 responses with custom instructions (00:24:36).
Dope, did you do this by hand?
😅😅😅😅
I was expecting "By Tammy AI"
Thanks
No one can explain a topic like Jeremy👍
The moment I got to know that you and Andrej weren't included in the Time's list, I realized that the people making such lists have no idea what they are doing. Loved the tutorial, thank you!
Hoping to look forward it.Ur fastai stable diffusion course was Perfect to the minute details
very easy to know, practical! thanks Jeremy
I found this video so useful that I felt compelled to pull my keyboard closer toward me, fix my posture, and write this comment - something I rarely do. I'm a professional data scientist hoping to push my company's GenAI agenda and this video makes me feel like I can actually do it! Thank you for so clearly encapsulating the state of LLMs. I'd learned many of these concepts before and this video is the glue that now holds it together.
The best "intro" and Guide I have seen on this. Appreciate it so much that you took the time to put this together and share this with us (FOR FREE!).
By far the most useful practical guide to LLM's by length. Thank you Jeremy!
Bravo. One of the best CZcams videos I've ever watched. Concise, entertaining, and chock full of useful insights.
Great! It allowed me to understand how LM thinks and why.
thankyou for the talk
So comprehensive. Perhaps the best introduction I have ever seen to the topic. Thanks so much.
This video landed up on my feed and out of curiosity I started watching and before I knew it had watched the entire video and taken copious amounts of notes too. One of the best videos I have ever watched!
You are literally changing lives, all for free. Thank you.
I'm really grateful how much people sharing their knowledge, can't imagine learning stuff for free. this means a lot for me.
This is golden summary of the state of the LLMs, Thank You
Brillant walk through. No hype. It is a real skill to explain complex topics is coherent way.
So many papers are being released, so it is important to have well-grounded information to understand LMs. Great delivery as always and practicable advice. Thank you.
Fabulous tour of key points. Fantastic job! Definitely going to recommend this to people wanting a gateway into llms.
Very much appreciated this consolidation of the main LLM coding concepts to-date. Thank you!!
I liked the video even before watching. Thanks Jeremy for your work, always learning from your content.
Thanks for all you do Jeremy. I have learned so many things watching youtube as well as the PDLC tutorials. Your expanations are on point.
Jeremy you're one of the most legit AI person out there. An enormous thank you for providing this and all your content. ❤
Kudos, such pleasurable 1 hour and 31 mins and 12 seconds.
Hi Jeremy, excellent walkthrough! This is truly helpful. Please keep them coming!!
Great video. Watched it in one sitting. It's very interesting and engaging, and does cover a lot of areas on LLM, different model, types, examples, uses cases, etc. I learned a lot and hopefully will go through the notebook in detail and adapt to my use cases. Thanks for making this.
We do not deserve you Jeremy! YOU ARE AN AMAZING TEACHER AND HUMAN BEING! Thanks, really, for all these beautiful lectures!!
oh man great video
There are hundreds of LLM tutorial coming out everyday, this is the one that I have been waiting for.
Thank you very much, Jeremy. Fascinating to see where we have come. The prose to SQL thing blew me out of the shoes. Can't wait to try this out by myself.
Thought provoking one code block at a time. As usual Jeremy the king
I have waited for months for a classification and evaluation from Jeremy. For me, this is by far the most comprehensive technical summary and evaluation available for someone who wants to delve deeper. It took me several weeks, if not months, to gain even a partial personal understanding of the current hype. Thank you, Jeremy, for all your good work!👍
One of the best and most educational videos I've seen on the subject. Thank you, Jeremy!
Came up in my feed. Thumbnail and title boring. By mistake I pressed play. But it was so interesting. I feel so enlightened after having been talked through this. Thanks for sharing this!
Thank you Jeremy for this introduction. It just answered many of my questions and affirmed some of my doubts about how many of the applications that use LLMs work today.
Thanks so much Jeremy, been following you since Kaggle's launch. Inspirational to see an Australian continue to kick ass as much as you have in your career.
This is great. I don't know how to say how much grateful I am for your video. Thank you and keep the great work!
Can’t stop watching over and over again! Thank you 🙏
Such a great article! I learned a lot from this video, such as how complicated systems can be put together using a stack of models, illustrated in the RAG to name an example. Jeremy, you are such a kind person to share this with the world.
What a great primer! Very much needed! Thanks as always Jeremy!
Always look forward Jeremy to explain this topic. Finally it is here. 😀
Keep on making videos man this was highly informative and my regards to being a person who was part of forming this architecture!
Never miss Jeremy's lectures....
Mr Howard never disappoints. Thanks a ton as usual Sir.
I can not emphasize how incredible this video was
Thanks Jeremy for another wonderful lecture! Much appreciated.
Jeremy, we need more videos on this topic! Thank you so much!
Hands down one of the best videos on LLMs on the internet.
Thank you Jeremy for all of your work and for sharing such quality videos. ❤
Thank you for sharing your experience and knowledge, Sir.
This is remarkable. Thanks for sharing this topic for us Jemery!
This is amazing, thanks so much for recording this and sharing it 👏
This is pure gold!!!!❤❤❤❤ I would have never found this on my own
Thank you Jeremy!! One of the most insightful and helpful vlog posts on the inner workings of LLMs... Top marks!!!
Thank you for creating this amazing talk around all the basics and applications with language models, this is really helpful!
A true legend! So far, I have not seen a better educator than Jeremy. His approach of teaching is what all schools and universities need! I am always interested to learn more, whenever I hear Jeremy. Thank you!
Wow, thank you!
A wealth of Information and knowledge in this video .. a master piece .. Thank you
Thank you for this. Couldn’t have asked for a better video.
Great course! Hello from Almaty Google developers community!
I feel like this video was made personally just for me. Amazing.
Thank you!🙏🏻😀
Brilliant presenter!!
The "wolf, goat and cabbage" riddle example is just awesome. Gotta use it to illustrate what LLMs can't do and why. Cheers for that :)
This is absolutely fabulous. Thank you!
Hey thank you for making these available for free. ❤
Damn I watched the whole video and didn't even realise that it was 1+ hr long! Thanks a lot for the great content!
Perfect description of Functions at 46:30!
Thanks...great summary....now i know the relatiinship between neural network parameters and vector DB's
Thank you for sharing all this knowledge.greatly appreciate it
This is a real gem. Reminds me of the authentic, high quality training material from Andrej Karpathy. Looking forward to future similar tutorials if you decide to make them! Thank you!
🎯 Key Takeaways for quick navigation:
00:00 🤖 Introduction to Language Models
10:27 🧠 Neural Network Basics
16:38 🚀 The Power of GPT-4
24:53 🌐 Limitations of Language Models
25:23 💡 Language model limitations:
31:32 📊 Advanced Data Analysis:
36:18 💰 OpenAI API Pricing:
39:19 🧩 Using OpenAI Functions:
46:40 🐍 Custom Code Interpreter:
51:13 🐍 Creating a Python code interpreter
53:39 💻 Running a language model on your own computer
55:01 🏎️ Choosing a GPU for language model work
56:15 🖥️ Options for renting GPU resources
57:57 💾 GPU memory size and optimization
59:20 📚 Using Transformers from Hugging Face
01:00:06 🏆 Evaluating and selecting the right model
01:14:12 📖 Retrieval augmented generation for answering questions
01:17:10 📚 Overview of using language models for document retrieval and question answering
01:20:35 💼 Private GPT models for document retrieval
01:21:03 🎯 Fine-tuning language models for specific tasks
01:25:15 📊 Building a language model for SQL generation
01:26:36 💻 Running language models on Macs
gpt plugin? :D dauymn son
@@plebmarv9668 it's tammy ai, a youtube video talking points extractor
This is so well done and presented. Thank you.
Impressive video; I spent days learning these concepts on my own. Had this been released two months ago, it would've been a game-changer. Excellent summary.
Luckily i'm a few weeks behind you! Happy learning mate!
People like you change the world!
Such an excellent video, thanks for taking up your time producing this for those very much at the start of this journey
I think the major problem is the retrieval. Would love a video just on that (best practices, best models out there etc.).
Great video, came across on x and subbed immediatley.
this changed my mind on how AI works! Thanks alot !!
Happy birthday Jeremy! Just got to the section where your bday is revealed and it is today! Thank you for all the great work :)
I really enjoyed this talk. Thank you so much.
Great content, and I love the way you structured it. Thanks :)
Awesome stuff, always like learning from your videos. Been watching since FastAI v1.
Jeremy this is a gem of a video. thanks again.
Thank you! Amazing tutorial. Thanks
Thank you for the video, it was fantastic!
incredible, value-packed, practical video for developers working with LLMs.
Glad you liked it!
What a fantastic video. Really enjoyed thank you!!
I was waiting for this for a long time! Thank you very much
This is amazing and serious content. I love it. Thank you for making it available, greetings from Switzerland
always good to learn from Jeremy!
Another awesome video from the GOAT.
Excellent tutorial🙏 respect🙇♂️
Very good which is very classical of Mr Howard.
Excellent. Thank you for sharing.
Wonderful overview, gives me confidence to dive in!
Thank you so much for making this excellent teaching video
Super informative video. Thanks for putting this out!
What a very useful and informative video -- I watched this over the course of a day and took notes -- Thanks!
Glad it was helpful!
Jeremy, Congrats on the 100k subscribers.
Well deserved and hopefully a catalyst to get your invaluable content more exposure.