literally i've searched so many articles blogs and youtube videos to understand transformer from the core basic. And this video come up with all the stuffs , and i must say best video everr!
Hey Ajay, I've watched numerous videos on transformers, but yours stands out as the best-super clear and easy to grasp! Your explanations are fantastic. Could you consider covering transformers for time series in your future videos? That'd be incredibly helpful! Thanks for the great content!
The Best explanation on transfer with code, even tough I am a AI developer and know these things pretty well I watched the full video Love from Banglore
27:40 isn't the key, query and value vectors computed using weight matrices for each of them which have learnable parameters. The input vector is multiplied by weight matrix corresponding to each Q, V and K.
Wow, when notifications come to me I could not belived after 6 month, the great man produced the new one. you are a greate one who I am following on youtube. I have suggestion again please do a series on how to use transformers on time series data including multivariate on anomaly detection and prdiction task. Please
Thanks sir, me and my classmates from Turkey send you a warm welcome. We are waiting, we completed your previews series about transformers. we learned a lot, just we will wait for your next tutorial on time series.@@CodeEmporium
Congrats for 100k.I have been following your channel for about 1 and half years now.Keep up the Good work.I was also asking for some advice.I did andrew ng ML,DL course,GANS and NLP course all from coursera.I implemented about 30 papers on convnets.about 10 on nlp and 15 in GANS .I realized there is a gap in knowledge as the papers can get into proofs which are math or stats heavy which can become hard to understand.i also followed the Karpathy nlp playlist which was great but now i feel like i hit a ceiling.I want to eventually become a MLE (currently a civil eng student).I have gone back to studying DSA because i dont know a recommended way of studying ML going forward.There are topics like optimization theory,game theory,info theory which have been recommended for study but i havent started because am in the middle of my semester.What do you recommend i do or what resources do you recommend i use.How I can reach out privately to talk more on this.
you are currently deep down in the tutorial hell. if you go on you'll never get out as theres always something you dont know. you are at a very good place right now, just start applying for a role you are intrested. you do not need to learn everything, you do not need to know everything. its pointless. noone does that. just focus on one subject/role and try to organize your time and resources to be the best for that role. game theory, information theory, etc are not really needed unless you intend on doing research on the domains involved. if thats the case, your best bet is to enroll in a Phd position and that should guide you properly. if not dont waste your time anymore. apply for a job and learn as you go on. Best of Luck
Appreciate the time you put into this! Refined my understanding of transformer models. I like this graphic you have at 1:19:00 do you have a file of that available somewhere?
When creating a copy for the residual, wouldn't it be necessary to have residual_x = x.clone()? Otherwise wouldn't it just be a reference copy? Also there is a permute statement used to swap the heads and sentences, but I didn't see a second permute statement to swap them back in the returned values tensor before reshaping.
It’s a bit funny that you speak perfect English with absolutely no accent, but you pronounce “matrix” wrong. It should be pronounced “may tricks”, not “mah tricks”.
![[ 100k Special ] Transformers: Zero to Hero](/img/logo.svg){kind=logo}
