At 46:50 that is not bigram count table at down rather those are counts resembling smoothed probabilities. Bigram counts won’t come down due to smoothing. I I count cannot be 3.8, probability adjusted count can become float.

DONALD TEUMPS DEATH is correct

Thank you for your wonderful lectures. It is not clear to me what is the definition of the proximity function phi(x,y)?

GOD , all my doubts are cleared. Struggled for months and here it comes to end. Thank a lot.

Amazing! ALL the stuff of deep learning and graph, fine-tuning, alignment, transformer architecture, etc IS CLUBBED here, Love it!

At time 59:38 why you have only one positive sample int eh log likelihood and k negative samples. I thought there were the sane number of positive and negative samples "k".

This is an awesome course, started as a KZbin suggestion now binge-watching whole series. Thanks for all the handwork to all instructors

You show the equation C^* = (c+1) N_c+1/N_c. Where did this formula come from? what is the explanation for it? any mathematical proof or at least any intuition for it?

Excellent courses! It would be best if you considered repeating the questions asked. They are inaudible when coming from students.

Any Diffrence Between this course and previously completed course this channel on large language model ?

are these the same videos as the LLM playlist in this channel.

Thanks for bringing this course!!

That's the course I need also following LLM lecture playlist

@30:42, the definition of monotonicity seems to be incorrect.

really enjoyed this lecture. Had an idea of perplexity before but never got its intuition this clearly

that's great ❤️

As usual awesome content and tshirt ! thanks Yatin :D.

amazing

love Yatin's t-shirts :D

VERY VERY GOOD. THANKS!!!

Thank you so much for this wonderful series !!

Sir Can you list some good NLP Courses in description or pinned comments

I'm so thankful for these lectures 🙏

👍🏻

Nice lec sir , There is a lack of such quality content on llms ❤

Thankyou Sir

Great! There should be a few lectures that include practical implementations.

In palm ,how does palm model knows when to use python interperter for calculation and when to use only llms?

very helpfull lecture

thank you sir for making this course in public

beautiful explantation of the concept of Perplexity. Thank u so much prof for enlightening me

Sir being from a tier 3 college this series is a real gem for a student like me. I am still in my 5th sem Btech in cse but this pg level course is so well taught i really liked the playlist hatsoff for ur efforts please i beg u to continue more concepts and not to stop

Is there an error on this slide? 17:19 • Bigram: P(begun | has) - „begun” after „has”. • Trigram: P(begun | season has) - „begun” after „season has”.

I never commented on any video, this is my first comment, I am working as AI Engineer currently, thought of refreshing some basics, and started watching Retrieval-based Language Models-I lec 16, I felt it's worth watching complete playlist. whenever I watch some recent technology or subject content on youtube from stanford, used to feel like when will our IIT's teach such latest industry ready techology right in the college. But now I can see that happening, I am heartfully thanking you guys for making it public and accessible on youtube. India needs more such kind of courses on latest trends and techonlogy. I never did it, but I bow to your team of teachers 🙏....I beg you to continue such work, hope other branches follow your path of teaching industry ready/required skills&topics to UG/PG students(btw I am mechanical engineer enjoying working AI ML domain, that's why this last point 😉)

please continue making these types of videos please

very ossm exllent lacture sir

Thank you Sir for sharing these videos.Really helpful.

What are the pre_requisite to study the course??

Professor what are the prerequisite to follow the whole playlist

Such amazing quality content, On latest technologies, That's why they are IIT profs

how the router picks an expert in case all experts have same score ?

In regular transformer, each token had positionwise FFN. In the presentation, every token is seeing same set of FFNs, why is this? Shouldn't it be FFN1_0, FFN2_0,.. for 1st token and FFN1_n,FFN2_n,... for nth token?

is flash attention of any use in case of edge devices?

how is the window size determined?

pls elaborate the autoregression part, specifically how the i/p tokens are fed to decoder . Let say decoder takes 512 input tokens, at first only <SOS> token is sent, but what about rest 511 tokens?

Great

How to get access to the PPT??

How can I download the slides for this video?

Sir still same issues persits in google trasnlators, I checked He is docter and She is docter

❤❤❤