![Bitswired](/img/default-banner.jpg)
- 99
- 60 827
Bitswired
เข้าร่วมเมื่อ 29 มิ.ย. 2021
For business inquiries, reach me at: jimi.vaubien@bitswired.com
🦙 LLAMA 3.1 secret sauce: Simplicity & Data Quality
Hey there, today we're diving into LLAMA 3.1, Meta's latest open-source AI models.
I'll break down why this model is such a big deal, from its massive 405B parameters to its clever focus on simplicity and data quality.
We'll check out the main features, benchmarks, design choices, multimodality implementation and legal updates.
Let's go!
🌐 Visit my blog at: www.bitswired.com
📩 Subscribe to the newsletter: newsletter.bitswired.com/
🔗 Socials:
LinkedIn: www.linkedin.com/in/jimi-vaubien/
Twitter: bitswired
Instagram: bitswired
TikTok: www.tiktok.com/@bitswired
00:00 Quick Glance at Llama 3.1
01:14 Benchmarks
02:20 Design Choices: Simplicity
03:56 Design Choices:Data Quality
04:46 Multimodality
06:06 Legal
07:00 TLDR
I'll break down why this model is such a big deal, from its massive 405B parameters to its clever focus on simplicity and data quality.
We'll check out the main features, benchmarks, design choices, multimodality implementation and legal updates.
Let's go!
🌐 Visit my blog at: www.bitswired.com
📩 Subscribe to the newsletter: newsletter.bitswired.com/
🔗 Socials:
LinkedIn: www.linkedin.com/in/jimi-vaubien/
Twitter: bitswired
Instagram: bitswired
TikTok: www.tiktok.com/@bitswired
00:00 Quick Glance at Llama 3.1
01:14 Benchmarks
02:20 Design Choices: Simplicity
03:56 Design Choices:Data Quality
04:46 Multimodality
06:06 Legal
07:00 TLDR
มุมมอง: 178
วีดีโอ
🧠 GenAI Design Patterns: Map Reduce
มุมมอง 640หลายเดือนก่อน
Unlock AI's Full Potential: Process Unlimited Text! In this video you'll learnhow to: - Overcome LLM context limitations - Handle dense information in long documents - Implement Map-Reduce with Python & Claude API (Anthropic) 🌐 Visit my blog at: www.bitswired.com 📩 Subscribe to the newsletter: newsletter.bitswired.com/ 🔗 Socials: LinkedIn: www.linkedin.com/in/jimi-vaubien/ Twitter: ...
🐍Make your Python code MUCH CLEARER with this feature
มุมมอง 382หลายเดือนก่อน
Master Python's Structural Pattern Matching! In this video you'll learn to: - Simplify complex conditionals - Handle intricate data structures effortlessly - Write cleaner, more expressive code 🌐 Visit my blog at: www.bitswired.com 📩 Subscribe to the newsletter: newsletter.bitswired.com/ 🔗 Socials: LinkedIn: www.linkedin.com/in/jimi-vaubien/ Twitter: bitswired Instagram: instagram.c...
⚡️30x FASTER JS formatting & linting with a SINGLE tool
มุมมอง 2Kหลายเดือนก่อน
Discover how Rust is revolutionizing JavaScript development! This video explores the game-changing impact of BiomeJS, a Rust-powered tool for code formatting and linting. Time to replace Prettier & ESLint ;) BiomeJS: biomejs.dev/ 🌐 Visit my blog at: www.bitswired.com 📩 Subscribe to the newsletter: newsletter.bitswired.com/ 🔗 Socials: LinkedIn: www.linkedin.com/in/jimi-vaubien/ Twitter: twitter....
💊 Run Pandas 150x FASTER without changing a SINGLE line of code
มุมมอง 3002 หลายเดือนก่อน
Boost your data science workflow by up to more than 100 times without changing a single line of code! In this video, we explore how cuDF, a GPU-accelerated DataFrame library, can dramatically speed up your Pandas operations. Link to the Colab Notebook: colab.research.google.com/drive/1P4PWtpS8vhE_gKwQyyxjoSuai_hCefub?usp=sharing 🌐 Visit my blog at: www.bitswired.com 📩 Subscribe to the newslette...
🏆 Rust is CHANGING Python for the better
มุมมอง 3.1K2 หลายเดือนก่อน
Discover how Rust is revolutionizing Python at three levels: 1. Tooling: Rust-based tools for faster formatting, linting, and package management 2. Libraries: High-performance Rust-powered libraries for data manipulation, text processing, and JSON serialization 3. Runtime: RustPython - an alternative Python runtime with interactive shell, WebAssembly support, and experimental JIT compiler Learn...
💡 There is a SMARTER way to split your documents for GenAI apps
มุมมอง 2.2K3 หลายเดือนก่อน
Learn semantic splitting in this hands-on tutorial to improve your language model's performance on document processing tasks. We dive into a practical Python implementation for finding optimal segmentation points by meaning, essential for retrieval-augmented generation. Code along with me following the GitHub-hosted notebook and elevate your app's efficiency with this smart splitting strategy. ...
🦾 The Cleanest way to write GenAI applications (it's NOT Langchain)
มุมมอง 4K3 หลายเดือนก่อน
In this tutorial, I show you how to easily integrate Large Language Models (LLMs) into your Python code using Magentic. It explores the powerful features of Magentic, including the @prompt and @chatprompt decorators, structured output, function calling, asynchronous execution, and streaming. You'll learn how to create complex LLM-powered applications with minimal boilerplate code, and leverage ...
🐍 3 Python debugging tricks you NEED to know
มุมมอง 1765 หลายเดือนก่อน
Debugging Python scripts can be a chore, but it doesn't have to be! In this video 👀 I’ll show you three effortless tips to supercharge your debugging arsenal, no IDE needed. 🌐 Visit my blog at: www.bitswired.com 📩 Subscribe to the newsletter: newsletter.bitswired.com/ 🔗 Socials: LinkedIn: www.linkedin.com/in/jimi-vaubien/ Twitter: bitswired Instagram: bitswired TikTok:...
🫵🏽 RAG or Tuning? make the RIGHT choice
มุมมอง 1785 หลายเดือนก่อน
RAG or Tune? That is the question! In this video we discuss the key differences between Retrieval Augmented Generation (RAG) and fine-tuning for customizing language models in just 3 minutes. Remember to subscribe to stay updated with more programming hacks, AI tips, and tricks. 🌐 Visit my blog at: www.bitswired.com 📩 Subscribe to the newsletter: newsletter.bitswired.com/ 🔗 Socials: LinkedIn: w...
🚫 Don't limit your app to ONE AI Provider
มุมมอง 7785 หลายเดือนก่อน
Looking to build a more flexible AI application that doesn't rely on a single provider? In this video, I dive into how you can achieve this with a Python library - LiteLLM. We will review the unified API and efficient function calling capabilities, plus how you can use LiteLLM's stunning features like observability, asynchronous support, and even its proxy server mode to leverage your AI applic...
🤯 AI text streaming with HTMX (no JS)
มุมมอง 1K6 หลายเดือนก่อน
Looking to master AI text streaming with HTMX, (without writing a single line of JavaScript)? In this video I break down how to set up a uni-directional SSE (Server Sent Events) endpoint and infuse HTML templates with reactivity using HTMX. I will also share a workaround to prevent unwanted browser reconnection after the SSE connection ends. This is the stack powering RustGPT, a ChatGPT clone I...
🧠 AI vs Machine Learning ? in 3 Minutes
มุมมอง 1199 หลายเดือนก่อน
In today's video, suggested by @pmarcum7146 (thank you!), I'm demystifying AI and Machine Learning for you in just 3 minutes! I've got examples like IBM’s Deep Blue and Google's AlphaGo to help understand these concepts better. Remember, AI is about creating machines that mimic human intelligence, while Machine Learning is a specific technique used in AI. Find out why Machine Learning has becom...
🧑🏽💻 I Built a README Generator with AI!
มุมมอง 1.8K9 หลายเดือนก่อน
I take you through the creation process of a new AI tool feature with LazyCodr using prompt-chaining. LazyCodr a Python-written CLI that simplifies the process of generating README.md files for git repositories and I will be guiding you through the challenges I faced and the solutions I discovered along the way. I will also address concerns about handling large files within the code base and in...
📊 REVAMP Your AI App: Visualize and TUNE Your Semantic Cache
มุมมอง 547ปีที่แล้ว
Hey there! Welcome to our TH-cam deep-dive into semantic caching. Looking to save on your OpenAI API costs and speed up your AI app? You're in the right spot! We kick off by showing how semantic caching, our high-speed data storage hero, outshines traditional caching in AI applications. Then, we get hands-on, building a web application with Streamlit to populate a cache and visualize results. W...
👌🏽 AI Chat Cheaper & Faster with Semantic Caching
มุมมอง 1.2Kปีที่แล้ว
👌🏽 AI Chat Cheaper & Faster with Semantic Caching
🤯 AI Writes Your Pull Request (LazyCodr) !?
มุมมอง 809ปีที่แล้ว
🤯 AI Writes Your Pull Request (LazyCodr) !?
🧠 Turn Websites into Powerful Chatbots with LangChain And Chroma
มุมมอง 8Kปีที่แล้ว
🧠 Turn Websites into Powerful Chatbots with LangChain And Chroma
🎙️ Google Sheets + Whisper: Effortless Audio Transcription!
มุมมอง 1.7Kปีที่แล้ว
🎙️ Google Sheets Whisper: Effortless Audio Transcription!
🎨 Google Sheets + AI: Instantly Generate Any Images! (DALLE)
มุมมอง 2.4Kปีที่แล้ว
🎨 Google Sheets AI: Instantly Generate Any Images! (DALLE)
📱 Boost Your Social Media Content with Google Sheets & GPT-3!
มุมมอง 655ปีที่แล้ว
📱 Boost Your Social Media Content with Google Sheets & GPT-3!
⏱️ Machine Learning Explained in 3 Minutes: Quick & Simple!
มุมมอง 3862 ปีที่แล้ว
⏱️ Machine Learning Explained in 3 Minutes: Quick & Simple!
🛠️ Top 5 Essential Tools for Every Data Scientist
มุมมอง 3712 ปีที่แล้ว
🛠️ Top 5 Essential Tools for Every Data Scientist
🔍 Debug ML With Overfitting: PyTorch Lightning (Tutorial + Example)
มุมมอง 8582 ปีที่แล้ว
🔍 Debug ML With Overfitting: PyTorch Lightning (Tutorial Example)
🚀 TorchData 101: Master Data Loading in PyTorch Like a Pro!
มุมมอง 2.1K2 ปีที่แล้ว
🚀 TorchData 101: Master Data Loading in PyTorch Like a Pro!
📰 Bitsletter #1: AI Drawing, PyTorch & Lightning Updates, React 18.0
มุมมอง 4582 ปีที่แล้ว
📰 Bitsletter #1: AI Drawing, PyTorch & Lightning Updates, React 18.0
💡 Cardano Smart Contracts: Plutus Pioneer Program Introduction
มุมมอง 4702 ปีที่แล้ว
💡 Cardano Smart Contracts: Plutus Pioneer Program Introduction
I like your videos 👍❤
Thank you 🙌🏼
I've always mocked Node for the node_modules being several megabytes or some gigabytes in size... Until I started using Rust... And my "target" folder being 2.4GB on an empty project with only deps :x I love Rust though and recently I took Typescript and I'm loving Svelte and Sveltekit :3
Ahah definitely! The cool thing with Rust though is that you get a standalone executable at the end, pretty easy to deploy to a barebone Docker image (contrary to other languages where you link to 3rd party libraries that must be installed on the machine). Nice choice, I love Svelte and Sveltekit too! Even though I have more experience with NextJS and React, the Svelte approach to reactivity is refreshing 👍🏽 What are you building?
The video is very good, but the background music is annoying for my taste, maybe low its volume?
Thank you! I’m glad if you learnt something :) Sorry for the background music, I try to not make that mistake from now on 😅
This is VERY useful. Thanks
Glad it was helpful!
Awesome breakdown! 🤯 LLAMA 3.1 is a game-changer in the AI world with its staggering 405B parameters. Can't wait to see how the focus on multimodality enhances real-world applications.
Thanks frero ❤️ New video on multi modality when it will become available 😁
Your channel’s evolution is insane 😍🔥 Keep going !🦍💪🏼
Thanks Mims ❤️ 🦍🐆 EKIP
Hlo bro ❤ collaborate karege mere sath
?
@@bitswired what happen
I did not understand your message? 😅
@@bitswired okay 👍 I want to collaborate with you
Thanks!!!!!!!!!!!!!!!!!!!!!!!
My pleasure ;)
This is essentially a part of what RAPTOR does. It's just that this is just to get a summarized document, where RAPTOR is used for RAGs. But the principle of chunking a document, summarizing it and aggregating the answers is similar between both concepts.
Thanks for sharing, I did not know about RAPTOR. I find the idea interesting, it could deserve its own video 😁.
Looking forward to crazy recursive functions that use these features
Ahah definitely I love Rust pattern matching for that Now we have something similar in Python 💪🏽
What presentation software are you using to show those kernel windows and chart divergence? (Great work btw)
Thank you! I’m glad you liked it :) I used Canva to build the animations and then I played them during the video
SIUUUU
Yung Padre for the intro ;)
Banger again!! Keep it up bro
Merci frero ❤️ Let’s goooo!!
🔥🔥🔥
❤️🦦
Gonna try this for sure keep it up bro !
Thanks frero 💚 You can try it in your Django project ;)
My favorite engineer ❤
Le plus chanceux ;) 🦍🐆
thank you very much
My pleasure ;)
I want to point out, with this tool it's not a magic that "node_modules" only have to install few things. If we take a look into `biomejs/biome/blob/main/Cargo.toml` file, there's a ton of dependencies for Rust. And then also take a look into `/biomejs/biome/tree/main/crates`. This is the price of making "node_modules" cleaner: all those supported languages, linter rules and other things aren't going to magically disappear, they still need to be downloaded, compiled, and executed. So I wouldn't call it "replace everything with a single tool", but rather "replace your JS ecosystem with Rust". I'm not sure how would I feel about that, given I don't really have Rust knowledge, but to me it seems a bit fishy, because now you'd have to convince people to add support for building Rust apps to your containers. While I appreciate an effort of making just a single tool for everything, the downside is added responsibility for introducing an entirely different ecosystem to your build process.
Hey, You have a fair point. Personally I'm comfortable bringing BiomeJS in a JavaScript project since Rust is abstracted away: you don't even have to know it's written in Rust, you can just run it as any node package. Also you're right that it's not magic and Biome needs to re-implement all these tools. But from the JS developper perspective it's cool to have a single tool to install and manage (upgrading versions and so on ...). But I totally understand your concerns. What is your typical stack for code quality?
Thanks for the suggestion, I'm going to try it now. Can you share what tools you are using for making such smooth videos? The smooth zooms and transition and stuff.
Thanks! Let me know your thoughts :) I use a combination of - Tella TV to record screen and camera at the same time with different layouts - Canva to create the graphics - Davinci Resolve to assemble the video, process sound, color correction …
Great job bro! Super good tool, gonna try this now :D
Thanks frero ❤️ let me know if you try :)
Imma try this now
Cool! Let me know your first impressions?
Really interesting. Keep on providing this advising stuff !
Thanks boss 🫡 I won’t stop 😉
Always a great content, keep going on ! 🦍😎😍
Thanks Bella ❤️ road to 1000 🦍🐆
hello
thanks
Keep it up bro
Thanks man! I will 💪🏽
Thanks for the quality of video, really interesting subject well exaplained
Thanks bro ❤️ Soon Elden Ring ahah ;)
Grat Video! But totally annoying music
It makes is hard to understand you and it distracts from your great work
If your video content would not be so great, I would have stopped watching
Thank you for the feedback! I reduced the video background music in the new video. Hope it’s better :)
Once all the vectors are loaded into the vector database the text splitting no longer matters. As long as you dont split on a compound word or phrase it doesnt really affect the vectorspace.
Hey :) I see your point but I would say that in practice it’s not the case. For instance if you embed an entire page versus multiple smaller paragraphs the resulting vectors will be different even though you’ve indexed the same text. And it affects the similarity search. That’s why pyramidal embeddings are a way to improve RAG performance by indexing the data at different precision levels and using multiple index to answer queries.
How is parallel programming done in Rust? I know that there are several MPI libraries such as OpenMPI, MPICH, Intel MPI, etc for running a C/C++ code (which uses these libraries) on multiple compute nodes (each with tens of cores) but what about Rust?
Hey good questions! On single node using multiple threads Rust has native support for real multi threading, and you have libraries like rayon that provide nice utilities. For multi nodes with Rust I’m not an expert but there also exist MPI bindings, like rsmpi.
You forgot to mention rust takes 20 years to compile lmao great video tho man thanks for the info. I work in Python and it sucks hard. These tools you mentioned look great :) Also could you include links to the tools and libraries next time?
😂 I had to be quiet about it 🤫 Just kidding, it’s a good point I forgot to mention! Thanks! Let me me update the description with the link 👍🏽 Hope you’ll like them ;)
Ah man I was really looking forward to a Mojo shout out 🥺the people creating it are amazing, and what they're doing with the language design of Python is simply jaw-dropping Definitely something you should consider for an upcoming video, subscribed! 😁
Wow man it’s a nice catch! I have to admit that I lost track of the Mojo project after they announced it a while ago … Thanks for the idea I will give it a try 👍🏽
Nice explanation !! Can we use gptcache for a finetuned codellama model?
Thanks! I’m glad you liked it Yes you can! GPTCache can be used for any model. It’s also integrated in langchain if you use it or you can run it as a docker container in server mode. Let me know if you have any issues ;)
Rust isn't even a real language.
It’s much more ahah
will it fix the syntax though?
Ahah it could If you make a pull request to RustPython you could change the language parser and have your own syntax Though, the chance to get accepted in main is low 😅
@@bitswired I'd be happy to see better support for Bython to be honest, sometimes the Py ecosystem has basically everything I need but even these few dozens of lines of glue in Py make me suffer
I did not know about Bython thanks for mentioning it! I don’t think it’s supported by RustPython but you could give them the idea 🤷🏽♂️ Ahah what are your favorite languages?
@@bitswired I spend most of time and feel most comfortable with PHP, JS and TS, but I can pick up basically any C-style language quickly, wouldn't want to bother with manual memory management though besides that I liked Delphi quite a lot, Lua was great for game scripting what annoys me in Python is the lack of braces (my brain spends excessive amounts of effort to properly track blocks without them) and the tendency to output integer results from math operations if you forgot to make all operands floats then there's AHK, I'm using it because there are no other options, but it feels like there's no rhythm nor rhyme how you do some things and every time I just shuffle things until it starts working, function calls there are bizarre
Cool I also use TS quite often 👍🏽 Yes auto memory management is a neat feature. You’re less likely to to shoot yourself in the foot 😅
I enjoyed the video, yes just need to work on that voice volume, and maybe a bit more upbeat voice. gj
I’m glad you liked it thanks for the feedback 😁 I will fix these and issues already working on the next video 💪🏽
Prototyping and overall development is faster in Rust than Python. Python will still be used because Rust has a steeper learning curve and a smaller ecosystem. But with time, Python will be replaced by Rust in many areas. Rust is "all levels", not just low level. Any python code can be translated to Rust code with zero level reduction. The borrow checker forces the Rust code to be "high level", either by using it or by avoiding it (and using smart pointers).
rewrite everything in rust!
Prototyping in python is way faster than rust… Rust will never replace python, but it will more likely replace C/C++ Rust is more similar to C/C++ than python. Rust and python were designed for two completely different use cases
Hey thanks for participating :) I see your point, and it’s true that if you are proficient in Rust you can go fast. But I think Python is still way faster for prototyping: - it’s not compiled so you can run a huge project instantly after making a change, while rust could take a while to re compile - the learning curve is way lower - the ecosystem is huge I also like your POV about rust being low-level and high-level depending on how you use it 👍🏽
Yes, I think it’s more of a c/cpp contender 👍🏽
Ahah hard to resist the appeal 😅
Clear, concise and very instructive. Keep up this great videos work.
Thanks Vogun ❤️ Let’s gooo ;)
Great job as always bro! Keep it up :D I might have to start my Rust journey looks like a god tier language
Thanks frero ❤️ Lets go you love Rust hehe After Transcendence ;)
Your videos are very well crafted, the only thing for me is the music maybe a bit quieter, it is a distraction for me from the point of the video itself (but this one is a LOOT better then the previous ones, there, i could not focus on the speech itself because the music did not serve as a background to well). Otherwise, keep up the great production quality, respect 👍
Thanks for the kind and useful feedback! Sorry for the distraction, I tried to make the background music lower and use a chiller song. But I will make it even more quiet next time 👍🏽
@@bitswired I agree, a somewhat lower volume would help. Great video.
Thanks for the feedback 👍🏽 I’ll fix the problem for the next video next week 😉
your audio sux then :P
You manage to make me understand an aera that is absolutely not mine, well done 😂🥰❤️
Thanks Bella ❤️ You will be soon an expert 😉
This is one of the most powerful videos related to AI I ever seen. Very clear, very informative, and very useful. Thanks for the good content 🌹🌹🌹
Thank you very much for your kind words! It means a lot to hear that the video had such a positive impact on you and it makes all the effort worth it. Thanks again for watching and for taking the time to leave such a thoughtful comment 👍🏽
Good presentation but I do not understand how it's different from document AI's that can do this automatically. Why do this manually?
Hey :) You’re right there are libraries that does it for you. However the purpose of the video was to understand how it works in depth, to do so I proposed a simple implementation from scratch. The goal was to help people grasp the concept. I hope you still enjoyed the video 😁
Very interesting
Thanks big boss ❤️
Love it ❤ You know how to transmit your passion, congrats 😍🦍🔥
Merci Bella ❤️🦍🐆 EKIP au max!
Great video bro, keep going with these fire topics!
Thanks frero 💪🏽 Let’s gooooo! Let’s make it work and play Elden Ring soon ahah
Slow but...
Yes depends on your machine and the mode your run With GPU enabled it should be better if you have access to one
Phidata does that a little better imo
Interesting, thanks for sharing I don’t know about phidata I will check it out 👍🏽
Great video…and very useful as usually with all of your contents. Keep on the great work!
Thanks Vogen 😉 I will do my best to be up to it 💪🏽
wow.. amazing. thanks for covering this
No problem :) Glad you liked it! What are you working on?
@@bitswired building my own agent platform. this is a perfect base to build it on. have my own twist on agents. lets see how it works
Congrats man! Wish you the best luck for your platform! Let me know when it’s up, I would happily try it :)