![Yujian Tang](/img/default-banner.jpg)
- 132
- 66 352
Yujian Tang
Приєднався 19 жов 2011
Llama 3.1 vs Llama 3 for RAG performance on Unstructured Data
Right after the release of Llama 3.1, Nina Lopatina from Unstructured has already done some evals. What's she learned? Come hear about her RAG evals on July 30th!
Follow for more events - lu.ma/oss4ai
Follow for more events - lu.ma/oss4ai
Переглядів: 314
Відео
Real-world Data Prep for LLMs: Challenges and Solutions
Переглядів 6146 днів тому
Building LLM applications? One of the top problems you'll face is going to be presenting the LLM with good input data. Good LLM responses need good input data. Clean, native text PDFs that are used in explainer articles and example code are rarely what you'll encounter in production use cases. Real-world data is wild to say the least! Here are some challenges you'll face: - Scanned PDFs - Sca...
Graph RAG Demo
Переглядів 997 днів тому
What does Graph RAG actually look like? Watch Jason Koo from Neo4j walk us through a Graph RAG example. For more tech talks like this check out lu.ma/oss4ai #genai #technology #neo4j
Make RAG better with Graphs
Переглядів 587 днів тому
You can use RAG to make LLMs better by injecting your data. You can use graphs to make your RAG apps even better via Graph RAG. Jason Koo from Neo4j explains how this works. For more tech talks, both online and in person, check out lu.ma/oss4ai #graphdatabase #neo4j #llms
What is RAG? Explained by Jason Koo
Переглядів 227 днів тому
Jason Koo explains RAG. For more tech talks like this check out lu.ma/oss4ai #ai #llms #machinelearning
Everything You Need to Know about Graph RAG
Переглядів 787 днів тому
What even is Graph RAG? Jason Koo from Neo4j breaks it down for us. Find more tech talks both online and in person at lu.ma/oss4ai #techtalktuesday #devtools #ai #graphdatabase
Graph RAG with Where's Waldo
Переглядів 597 днів тому
Jason Koo from Neo4j explains what Graph RAG is and how it works via Where's Waldo! Find more tech talks both online and in person at lu.ma/oss4ai #ai #graphdatabase #programming
How to Create Breakthroughs in Data Management
Переглядів 157 днів тому
An excerpt from my interview with Chang She, CEO and Co-Founder of LanceDB. Learn more about one of the most innovative databases, optimized for multimodal AI training, analytics, and retrieval, and up to 100 times faster than Parquet for many AI workloads. Keep up with more talks like this by subscribing to lu.ma/oss4ai #genai #devtools #programming #data #machinelearning
The CEO of a Vector DB Company Speaks on the Future of AI
Переглядів 137 днів тому
An excerpt from my interview with Chang She, CEO and Co-Founder of LanceDB. Learn more about one of the most innovative databases, optimized for multimodal AI training, analytics, and retrieval, and up to 100 times faster than Parquet for many AI workloads. Keep up with more talks like this by subscribing to lu.ma/oss4ai #genai #ai #devtools
How LanceDB tackles CAP Theorem
Переглядів 428 днів тому
An excerpt from my interview with Chang She, CEO and Co-Founder of LanceDB. Learn more about one of the most innovative databases, optimized for multimodal AI training, analytics, and retrieval, and up to 100 times faster than Parquet for many AI workloads. Keep up with more talks like this by subscribing to lu.ma/oss4ai #ai #devtools #programming #databaseconcepts
An innovative database changing the whole AI field : the idea behind LanceDB
Переглядів 348 днів тому
An excerpt from my interview with Chang She, CEO and Co-Founder of LanceDB. Learn more about one of the most innovative databases, optimized for multimodal AI training, analytics, and retrieval, and up to 100 times faster than Parquet for many AI workloads. Keep up with more talks like this by subscribing to lu.ma/oss4ai #genai #ai #devtools
Awesome AI Dev Tools by OSS4AI - July
Переглядів 17116 днів тому
This month's talks feature: - Video AI by Christian Safka and Ari Korin from Tavus - Should you learn an LLM Framework by Yi Ding - 14 startup pitches
Building LLM Evals That Work IRL
Переглядів 5617 днів тому
Arize AI is a leading company in the evaluation space. Not only have they been a player since before it was popular, but they have plenty of real world experience. Aparna Dhinakaran comes and talks with us about lessons learned. Clipped from the Mid Year Gen AI Zoo by OSS4AI. Want to learn more about #ai? Follow lu.ma/oss4ai for more information. #devtools #llms #generativeai
Manage, Visualize and Search Video Data on Your Laptop
Переглядів 2617 днів тому
Computer vision data is large by nature, and that makes it tough to manage on something like your laptop. Visual Layer has made that possible through their new open source tool - fastdup. Danny Bickson talks to us about how to use it. Clipped from the Mid Year Gen AI Zoo by OSS4AI. Want to learn more about #ai? Follow lu.ma/oss4ai for more information. #devtools #computervision #genai #artifici...
A Database for Multimodal AI,
Переглядів 2617 днів тому
As we've seen already this year, multimodal AI is top of mind for many AI practitioners. Vishakha Gupta has done research on AI at Intel and brought her expertise into the startup world through ApertureData, a vector graph database. Clipped from the Mid Year Gen AI Zoo by OSS4AI. Want to learn more about #ai? Follow lu.ma/oss4ai for more information. #generativeai #devtools #llm #llmops
appreciate it!
RuntimeError: Trying to backward through the graph a second time (or directly access saved tensors after they have already been freed). Saved intermediate values of the graph are freed when you call .backward() or autograd.grad(). Specify retain_graph=True if you need to backward through the graph a second time or if you need to access saved tensors after calling backward. I took this error. (I am not using the same dataset). Thank you.
😮
Haha we have actually made an app similar to Parallel live (with agents) for a hackathon
thank u!
I have two questions : 1) I understand that this is multiplication free, with ternary weights being essentially additions and subtractions. However, it's still not clear to me if these additions and subtractions are performed on floating point numbers (presumably bf16). 2) I looked at the paper and I don't understand the relation between the MLGRU and the GLU. They seem to share a Wg matrix, but is it really the same ? Fig.2 in the article a matrix Wm is shown, but not in the equations of th eGLU. Very confused.
Looks like very interesting techniques. And great to see the team also looked at the potential implementation cost reductions from MatMult free. It will be interesting to see how this scales up to larger models.
Very cool, could you share the workflow and Notion notebook?
obviously you have to read the docstrings to do the Class LSTM(nn.module): part? ....no?
Ace is both highest and lowest in most games of poker and black jack.
Can you do high/low with suits and face_cards working with lists and tuples?
yeah
@@yujiantang I ended up figuring it out last night- took me a while to work out the game logic kinks, now I’m just working out game-flow. You helped changed my perspective though so thank you. Excited to see how you execute it
I know this an old video, would it be possible to change the code a bit to work with "Deck Builder" type game instead of just a standard deck
yeah, just replace the cards
Nice and simple I like it.
Nice and simple Like it.
1:00 Yess, exactly the point: long system prompt + input.
hi! nice content 🤝
Yeah, agreed. Need more of it
thanks!
Thank you very much for your work! Helped me a lot!
nice!!
Bruh you're quick with it, but also straightforward. I like your style of teaching for sure!
why not for i in range(0, number): if i * i == number: print(i) much easier and you get the value of what number is the square root
sure, you can do that, the runtime for that is a bit longer though, especially as your number gets larger
@@yujiantang if i'm not mistaken there's a module to remove caches and unnecessary codes to speed up fibonacci sequence, i think it would work on this too
What is the names of the algorithm used in here? Does it have a mathmatical name?
no idea really 🤣 this is the method I used to use when I did math contests, just implemented in Python
thank you for this! i'm working on a blackjack project for class and didn't know how to approach the deck of cards, this will help!
Sorry for misspelling your last name, @Yujian Tang
Hey Yujian, can we use a csv, instead of a json? To use as a file to store the information? @Yujian Jung
yeah, you totally can, that would be for more tabular data - ie you want to be able to related different entities and such
Really?
yes, this is really how gpt works (at a high level)
I managed to create a deck and to even shuffle the deck! Now I just need to figure out how to deal the cards out to multiple players and I'm one step closer to making a card game.
just checking in - how is that card game going?
I have a question regarding the step n**2 <= num ; so I understood that this means that the prime factors of a number all exist less than or equal the square root of a number num. But in the case of num as 28 , it has a prime factor of 7 , and 7 squared is not less than or equal 28 , but somehow the code also worked and i got 7 included in the prime factors as a result. So does this code only checks prime factors less than or equal to square root of a number ? or did i get this wrong as somehow it also worked correct with 28; maybe this while statement only means that for n less than or equal square root of the number to divide num by n??
the code will add both the smaller and larger factor when n==4, you got 7 back because 4 x 7 = 28
hi! great video! what extension do you use to color code everything?
haha i think it's dracula
if I do "while n <= num", I get the same result. Why should I do "while n**2 <= num"?
both will yield the same results, using n**2 uses less loops so it's faster/better/stronger/harder/etc
this is actually work! thank you!
well of course it actually works :p
recently came across this channel and I'm enjoying this style of content, many thanks!
Right now doing an assessment, this video really helped out.
Don't you use elif and else instead of 3 if statements?
Write a program by using class and methods to show winner if below scenario pass in deck of cards. King of Spade, Queen of Heart King of Heart, Queen of Diamond King of Diamond, Queen of Club King of Club, Queen of Spade Hint: Input should be two digit like (2,5), if 2= King of Spade and 5= Queen of Heart then person is the winner. can u help me with this scenario
cool
there is not enough link to githov so as not to copy the code manually from the screen to your computer. think about it in the future. the video also starts with some kind of vague start, as if I had to watch some previous videos before it, although it does not say that this is part 2 or something like that.
Here's the github: github.com/ytang07/nn_examples/blob/main/pytorch/rnn/lstm.ipynb
you are the best teacher I have ever seen !
Why do we need to intialize the hidden_states and cell_states of the lstm every training step?
hidden state weights are not retained from each training step, also i don't believe i'm initializing them every step, just each epoch, you can also re-initialize each batch if you are doing truncated backprop through time
nice, short and clear
Yo, dude keep going, keep creating videos like this i want to learn from you
Thanks! Glad that these are helpful :)
*promo sm* 💥
First
Very nice video, I have a question about PCA If you train a model, let's say EigenFace, in OpenCV, and you want to use it to compare the trained model with an image, using PCA, how can you do that? Basically my doubt is comparing an Eigenfaces model with an image and the output should be the closes PCA image and its class
Hi Gianmarco, I'm not sure I totally understand the question, what do you mean compare the trained model with an image using PCA? You want to compare the eigenvectors?
@@yujiantang yes, compare a trained model with an image based on the eigenvalues and select the image that is the closest to the new input
Its cool seeing someone use flask for a change! Cant wait for part 2 :)
Thanks! It's up today :)
Thanks for the vid good help for us beginners. Can I ask what is the empty card list for? Thanks!
Hi Benedict, great question. It's so that we have an object to insert the cards that we are creating into!
how to do with a function?
Hi Ana, you could put all the code into a function and that should do the trick, if you have more questions or want to talk about the code itself, feel free to jump into here --> discord.gg/ysARSHdkcF and ping me
Thanks alot!
Bruh. This is exactly what I was looking for. You the realist. Just found your channel and will be munching on your videos until finals are done.
Awesome to hear Shawn, happy to help!
lol you type so fast
lol thanks! i developed my typing speed from playing too much league of legends as a high schooler 🤣
Short & simple. Thanks
Awesome, glad to be of help!