$0 Embeddings (OpenAI vs. free & open source)

[@phobosmoon4643]

45 seconds in and you have already asked all of the correct questions. you have my attention, sir.

[@kop-lg7lo]

And my like

[@ssathessa]

16:00 in and you have completely lost my attention, along with my hopes for being to do this as a 1st timer😂

[@funkdefied1]

@@kop-lg7loand my comment

[@TC-Loom]

After watching dozens of hours of similar videos this year, this is the best one. Thank you

[@carpal4489]

ah yes javascript and co way better huh?

[@erikmejiass]

I don’t usually post comments, but in this case would do an exception to thank you for the easy to understand and to the point explanation. It’s rarely found these days. Hats off to you 💪

[@RabbitHoleSyndrome]

Thanks! Glad it was helpful!

[@eyemazed]

damn i love when i dive head into NLP content, get instantly overwhelmed and then a video like this gets randomly recommended and instantly clears away about 120 questions that i had. thanks man

[@RabbitHoleSyndrome]

Awesome!

[@Richardziani]

Hey, I totally feel you! 😍 It's amazing how stumbling upon the perfect video can make all those overwhelming questions disappear in an instant! 🙌 Thank you so much for sharing your experience, it's truly inspiring! 🌟 Keep diving headfirst into NLP content, my friend! 💪

[@Mullheimer]

It's amazing how Google just reads your mind. At this point it is just getting retarded. Every day exactly the right yt vid comes up. I've been querying large pdfs today via chatgpt...

[@justinyoung1762]

This is the best ML for JS tutorial I've seen. SUPER helpful that you started with foundational topics. Thanks for making this.

[@mortocks]

First tutorial that actually makes sense and explains the ecosystem without hand waving. Thanks!!!!!!

[@RabbitHoleSyndrome]

You bet! Thanks for watching 😃

[@kiiikoooPT]

that is actually not true, all the basic scripts in this video start with a hand waving instead of an "Hello World"

[@mohamedamine_47]

While I didn't originally come to watch the TypeScript implementation, I found myself fully immersed in your explanations. It wasn't until I'd watched the first 30 minutes that I realized the video was over an hour long. Thank you for sharing this!

[@fille.imgnry]

This is sooo good. The way your thinking, explaining, digging deeper, wanting to understand why things are happening the way they do. Thank you!

[@MadeInJack]

Man, you are the reason why KZbin is such an awesome platform. Thank you so much for this golden course!

[@comptvlee]

Insanely helpful content. You've been one of the content leaders carrying the burden of disseminating all these complex topics into my head. Started with the supabase docs all the way here and I'm grateful!!! This is game changing for my career

[@RabbitHoleSyndrome]

I appreciate the kind words. Glad they have been helpful!

[@jimg8296]

Wow that was comprehensive and informative. Never realized there was so much involved in selecting models for creating embeddings. Thank you so much!

[@johngrant7197]

I love that you rock Typescript in domains usually reserved for Python.

[@sheikhakbar2067]

It's almost 3:00 in the morning, I am going to sleep and watch this excellent video tomorrow!

[@JoEl-jx7dm]

Same but different day

[@SamirLohiya-sm7ze]

This is a great video, and it makes me feel like there is no need to refer to any other video. What a great piece of content!

[@zeelthumar]

I have never watched full tutorial that contains javascript bcz I'm a python guy but for the first time I have gone through the whole tutorial and obviously I subscribed...................U are just awesome...............keep it up

[@chrissparks8377]

With a heavy CompSci / Math background, I found this to deliver CRAZY VALUE ❤ in terms of knowledged gained. You touched on the data structures, algorithms, Web API's, JavaScript ecosystem trends, and most importantly, the fundamentals of text embeddings and modern trends around their usage and deployment in an approachable, hands-on manner. 🤯 I'm always on KZbin, but literally had to look for where I could comment -- this might be my 3rd ever... Fantastic work 👏 🎉

[@RabbitHoleSyndrome]

Glad it was helpful 😄

[@shabarinathk8954]

27:50 : Bro started counting from 0 , what a legend and a true engineer

[@richardyim8914]

This was crazyyy useful. Not a software or web developer, but trying to use embeddings for academic research. Really great walkthrough!

[@mahuzedada]

There's a tool that allows you to manage embedding without code

[@ninjagaskin]

This beats watching Netflix. Thank you for such a well-put-together tutorial/lecture on embeddings 101. I would really value your advice (and this community too) on a question I have for a personal project: I am attempting to build a pdf analyzer for financial reports - 10ks, Annual Reports, Financial Statements, etc... What is the best metric / category you recommend I use for my embedding model and LLM too? the goal is to allow the upload of any pdf, embed the pdf, store in a vector db, and then query some question answering along with memory capabilities. - new subscriber

[@RabbitHoleSyndrome]

Great use case! I would focus on a retrieval based model which is best suited for question-answer style matching. In terms of which one - I think you will just need to try some of them and test to see if your queries are matching with the content you expect. Good luck (thanks for the sub)!

[@scign]

I love how my head comes up with a question and you go down that exact rabbit hole! You just won a new subscriber here.

[@RabbitHoleSyndrome]

Nice! Thanks for the sub!

[@lofikiwii]

My man!!!! Huge! (it's tony), glad to see a new video! And this was one a monster video! Its jammed pack with so much information man. Keep it going my man!

[@cesar_ai_eng]

I've watched just two of your videos and I feel like I've learnt 6 months of ML Engineering. Many many thanks

[@RabbitHoleSyndrome]

Glad to help!

[@abagatelle]

I'm.....overwhelmed! Learn't that it's a very long road. Very pleased I found your excellent channel.

[@JohnnysaidWhat]

I used clip to embed some fields of study and then feed it user input which would "fuzzy map" to my embeddings. Worked so so well and saved me a ton of heartache. Didn't realize this model could do text2image and vice versa. Will have to try that out as well!

[@danielbusquets3282]

Very good video. I just implemented a semantic search engine in my app and it works like magic

[@M110ification]

This is the tutorial I've been trying to find. I like how you describe the different components of the system and didn't just start with a walk thru how to setup a dev environment and other mundane pieces.

[@pranjalagnihotri6072]

Thank you so so much for this banger video few days ago I was working on a side project (inspired by Wesbos talk) to find the similarity between a tweet replies and group them together I first used bag-of-words to generate word vectors and then used OpenAI embeddings only to find out I need to setup a paid account because of that I had to pause my side project and I would say right after this I am going and will complete my projects using huggingface models thank you so much. I explored HF but was not able to gather such details like you explained everything in video❤

[@RabbitHoleSyndrome]

Awesome! Glad it was helpful 😃

[@papa4614]

Yes very nice to see your still doing vids ❤❤❤ there is so much your teaching us, thanks for that. I hope you still do the part 3 video series on the electron screensaver project 🤩 stay save

[@Muennighoff]

Amazing video! Will work on adding a ranking for the individual MTEB task tabs. Also we're working on adding more tasks (e.g. code embeddings) & other models to the leaderboard! 🤗

[@RabbitHoleSyndrome]

That’s amazing Niklas, thanks for the update!

[@cklim78]

Great! Look forward to the updates especially on the code embeddings task, may I know any targeted date when the updates will be released?

[@superwin06]

Thank you for covering a lot of sub topics without going deep too much! I learned a lot

[@eru6ite]

Dang! I'm not even halfway and I've already learned so much. Amazing content, mate! 🍻

[@shuaiber]

Fantastic content. Loved the contextual information in the beginning which is relevant to understand all the different components in the ecosystem.

[@rickgeyer9685]

This is absolutely the best explanation of embeddings I have ever seen. Thanks so much for this excellent video!

[@yuri.caetano]

Bro your video is hell good edited and content is good, congratulations

[@bakistas20]

I love how detailed your tutorials are! Keep on

[@SantamChakraborty]

Thank you ! That was amazingly lucid and well put together. Keeping an eye out for your future videos.

[@AngelEduardoLopezZambrano]

This video got me into the rabbit hole of your channel! You just got a new subscriber. Excited to see what you'll come up with next. Particularly, I'd like to see how to fine tune text generation models in modest hardware using javascript. Thanks!

[@RabbitHoleSyndrome]

Will keep that in mind - thanks for the sub!

[@jervx829]

Sir, Thank you 🫡. I have difficulty focusing to videos like this and I watched many tutorials before and often got bored & ended up closing it. But your explanation was so good and I enjoy learning from you. Thank you for sharing your knowledge to us God bless you sir! I wished I discovered you earlier when I was still on college last year, my capstone project could've been different.

[@jervx829]

For those who trying on Nextjs 14 app router (server side). Don't use pnpm. Switch back to npm

[@rikvermeer1325]

Hmmz... finally a good overview on embeddings that is in depth AND understandable I will watch this multiple times

[@snsa_kscc]

Another banger, bro! You are on 🔥. I love the fact ML ecosystem is morphing with TS/JS web dev. Massive thanks, this was a joy to watch. Have a nice one.

[@RabbitHoleSyndrome]

Appreciate it & thanks for watching! Very excited to see the new possibilities in the TS/JS ecosystem.

[@stephenk8632]

Amazing video! Thanks. It was uncanny how questions would be popping in my head and you'd answer them in the next sentence lol. Thanks again

[@JunYamog]

Thanks I made it on the end of the rabbit hole. Interesting to learn that you can do embeddings in a browser.

[@neoblackcyptron]

I'm glad I found this channel. It's very rare to come across super knowledgeable people like the owner of this youtube channel in the AI field who really know their stuff. I am planning to start an AI consulting business, cutting costs reducing token costs is a big deal for me, this really helped. By the way I heard AI consultants who know their stuff: ML, Genetic Algorithms. Robotics, Computer Vision and LLMs with some RPA thrown in can charge $200/hour is that a fair rate or am I underselling me skills. I don't want to ruin it for the rest of my peer by quoting too low.

[@roguesherlock]

man this was so helpful. Thank you! Also, I loved all the btws, fyis, fun facts and side quests haha

[@stephensinclair276]

That was one of the best videos I have seen on this topic. Going to look for more!

[@martinp3839]

Excellent! Especially the way you build up the episode from explaining core concepts to a layman..glad I stumbled upon your channel..

[@RabbitHoleSyndrome]

Glad it was helpful!

[@martinp3839]

I am new to javascript and typescript. When I cloned the repo and tried executing "npm run dev" in the "HuggingFace/apps/embeddings-huggingface/src", get this error "TypeError [ERR_UNKNOWN_FILE_EXTENSION]: Unknown file extension ".ts"", Hit a roadblock. Kindly advise how I can fix this.

[@RabbitHoleSyndrome]

Sounds like you may be running the code as JavaScript (TypeScript has a compile step). The easiest way to get off the ground is probably to install ts-node and run your TS files through that.

[@ymroddi]

Just a side note for those using transformer models in production - interacting with the model should be done in a worker per official docs so it is off the main thread. I know your example was very bare bones so this wasn't an issue.

[@RabbitHoleSyndrome]

Great call!

[@shinchima]

just the ticket - cheers for uploading

[@browland601]

Love the level of detail here! I'm working on an idea to do text classification using sentence embeddings as input features and then doing logistical regression. If the labels (binary classifier for now) correspond well with the embeddings generated from training examples, then my hope is the logistical regression model will work with inputs not seen before but close semantically. One of the considerations is which embedding model to use, so lots of food for thought here!

[@theepicosityofpizza]

This is how they tested classification performance in the benchmark mentioned, so you're on the right track

[@omijmangukiya8984]

i suggest using setfit

[@hasaniqbal233]

Really appreciate you doing these videos using TS and not Python.

[@borknagarchile]

Your video is amazing, you pretty much see all the possible the use cases for when you are trying to implement something like this. Thanks 🎉

[@Bruh_688]

Love you man, thanks for coding and making this tutorial in JavaScript/TypeScript.

[@jagrat12354]

Every sec of the video was worth watching

[@PatrickSteil]

Wonderful video... helping to teach us how all this stuff is working under the covers... new subscriber!

[@crometica]

It's incredibly valuable, and you've gained a fan ❤. Keep up the fantastic work!

[@IgnoreMyChan]

I love your 2010-like editing with all the micro-cuts making it very exhausting to watch and listen to. 😂

[@moy2010]

Probably the best content I have watched on the subject. Kudos!

[@RabbitHoleSyndrome]

Happy to hear! Thanks for watching 😃

[@grant_vine]

Best hour and 24 mins spent this week 😊

[@khangvutien2538]

With this video, you got a new subscriber. Thanks.

[@vinitjha_]

Great explaination. You simplified every bit of detail👏

[@mytechnotalent]

Great job and very detailed I like the JS internals showing the steps.

[@RabbitHoleSyndrome]

Awesome!

[@ThunderboltPath]

I appreciate so much that you made a video about this as I’m learning. But more than 90 mins was really long and I had to skip over some detailed explanations to get to see the demo 😂 So if anything, I’d hope you did it in 2 parts, one video for just the demo of these new technologies (what they can do for me) and another going into details behind everything. That’s my 2 cents… Loved the content 🎉😊

[@luis96xd]

Fantastic video, everything was well explained! Thanks Wow, it was an excellent solution to get Array length 1 of embeddings forking the model and adding the label Sentence Transformer, you really research a lot before making a video, great job! 👏💯 I learned a lot, of programming of types, of model parameters and outputs

[@RabbitHoleSyndrome]

Glad it helped!

[@TheTaygan]

Great video with awesome explanations... well done

[@jakubwerner11]

So well explained, thank you!

[@aksh1618]

Another great one! I'm really hooked onto the high quality content you're producing! On a related note, which one of these open source embedding models would you consider using for something like ClippyGPT, and are actually you planning to switch to one of these?

[@RabbitHoleSyndrome]

Thanks for watching! Great question - we will definitely consider it. Will have to do a deeper dive, but ideally a model that performs well with retrieval. Also need to consider practicalities around deployment and operation, so usually smaller model = better.

[@danjsy]

Great video thank you an excellent channel. For corporate use cases where they are predominantly Microsoft houses, for retrieval use cases can you see past the new OpenAI implementations? Thanks !

[@Swanidhi]

Love the rabbit hole content! Subscribed!

[@RabbitHoleSyndrome]

Thanks for the sub!

[@gnomewc22]

Man ur just a natural teacher, thank you!

[@monterourena]

Amazing job! Please continue doing these videos 🙌🏻

[@eliaweiss1]

wow!!! amazing stuff, thanks for all this info!

[@handlez411]

Awesome video! Thank you!

[@SmallTownDev]

By far the most informative resource 🚀

[@Franx570]

I'm only 30 minutes in, but Jesus this video is gold 👏👏

[@shtrumj]

please more content , more depth, loved it, subscribed.

[@DavitBarbakadze]

Man I wish you'd create those videos more often. Maybe do a Patreon acoount. I'd gladly pay.

[@N4LNba777]

Amazing video, thanks for saving me a ton of time!

[@RishuKumar-zo2hc]

Very informative Please release more videos Love your content

[@sirishkumar-m5z]

An excellent comparison of embedding tools! Another free option worth exploring is Hugging Face's Transformers, which offers robust embedding qualifications. #AIEmbeddings #HuggingFace

[@lakinmohapatra]

Very nice explanation. Thanks a lot

[@jdkh89]

Wow I learned so much from this. Thank you so much!

[@RabbitHoleSyndrome]

Glad it was helpful!

[@hamzakyamanywa9792]

loved watching this!

[@gunerdown]

You got my sub, great content!

[@sarojregmi200]

I think they saw your video and added the sentence transformer label in there.

[@mohegyux4072]

lol I didn't know the name of the channel until minute 40, subscribed

[@RabbitHoleSyndrome]

Thanks for the sub! 😄

[@lucamiamiflorida5065]

best video on the topic, thanks!

[@RabbitHoleSyndrome]

Glad it was helpful!

[@manishbabbar9969]

Really liked your content. Thanks for sharing such engaging information.

[@RabbitHoleSyndrome]

Thanks, glad it helped!

[@KelvinWKiger]

Olala! Thank you, thank you... and thank you very much. Take care 🍀

[@justine_chang39]

this video is amazingly good

[@LWC]

Really great video. Thank you for it!

[@RabbitHoleSyndrome]

You bet!

[@artem-yw8km]

cool realy improved inference results

[@adrienazie]

Thanks for the time and effort you have put into creating this video. Great content and explanation.

[@aliandiazperez7602]

Excellent video tutorial and teaching skills!!!!!

[@RabbitHoleSyndrome]

Thank you! Cheers!

[@ahmedali-hy9ew]

Keep it up with the great content, really love the way you explain stuff, we want more ai tutorials please.

[@raphauy]

Thank you very much!

[@OG-ok9zh]

Did you have a chance to compare the performance regarding total request time vs execution time? Intuition tells me this could be MUCH faster than making a network call, server-side. Did you get the chance to verify this however?

[@RabbitHoleSyndrome]

Great question - no I haven’t tested this yet. Agreed on the intuition though - my gut is that network latency will almost always be greater than inference time. Also worth pointing out you have more control on availability (don’t have to worry about degraded third party API performance which I have experienced in the past)

[@OG-ok9zh]

@@RabbitHoleSyndrome I'll check it out and get back to you today then! Let's see!

[@eyemazed]

just a few questions though - are text embedding models language agnostic? and by language i mean the language text is written in, and if not, are there any benchmarks showcasing the performance? the second question is the token limit, is there a token limit for the embedding generation, in other words how big of a text can you provide to the embedding API?

[@artistry7919]

19:22 answers the second question

[@oscarljimenez5717]

You meant programming languages? If yes, well ML and AI are only math, so yes, i actually make my tesis on AI using Java at the time. You just need to understand the math and convert it to code.

[@RabbitHoleSyndrome]

W.r.t. languages - it will all depend on what languages the models were trained on. On the MTEB leaderboard you can see which languages were used (mostly English). You may have the best luck with the Bitext Mining models or models titled/tagged ‘multilingual’. Overall probably best to look into each model and their description on a case by case basis to understand which may be best for your use case (for now).

[@elierh442]

Love your content, keep it up! 👌

[@orenmizr]

that was very interesting. thanks🙏

[@RabbitHoleSyndrome]

Thanks for watching 😄

[@daryladhityahenry]

Hi! This is crazy good content. I learn so much! Thank you so much.. And have a question though, I'm on minute 39:00 when talking about embedding dimension size. Is the lower dimension size more likely to not have word vectored? I mean, maybe word like "eggyolk" ( maybe ) doesn't registered because the embedding id already full so it can't get the vector value from that word? If yes, that is really a trade off that we need to choose carefully right? Or... If we just use normal vocabulary, it should be okay with small dimension? Thankss!!!!!!!!