Generative AI (Advanced NLP) in Urdu, Spring 2026 (Agha Ali Raza)
By Agha Ali Raza - Youtube Channel
- Course audio
- Course audio: Urdu
- Level
- Advanced
- Length
- 31 lessons · 30.3 hours
What you'll be able to do
- Explain tokenisation methods such as BPE and WordPiece
- Understand word embeddings and their biases
- Follow how RNNs, LSTMs and attention lead to transformers
- Describe LoRA, quantisation, instruction tuning and RLHF
About this course
A recent, 31-lecture university course (about 30 hours) on how large language models actually work, taught in Urdu/Hindi. It moves from the evolution of NLP and tokenisation (BPE, WordPiece, unigram) through word embeddings, skip-gram, CBOW and GloVe, bias in embeddings, RNNs, backpropagation through time, LSTMs and cross-attention, up to modern topics such as LoRA, quantisation, instruction tuning, RLHF and PPO. It is heavy on concepts and maths, so it suits final-year students and developers who want to go beyond using ChatGPT to understanding the models behind it.
Before you start: Machine learning basics and some linear algebra
Syllabus
31 lessons · 30:15:20 in total
- 1Generative AI L1: Course basics, introduction to language44:59
- 2Generative AI L2: Evolution of Natural Language Processing1:15:57
- 3Generative AI L3: Scaling & computational realities, linguistic hierarchy, tokenization intuition1:17:23
- 4Generative AI L4: Types of tokenization (word level, character level, subword level), BPE algorithm1:13:03
- 5Generative AI L5: Tokenization (Word piece, unigram tokenization), language related terminologies59:34
- 6Generative AI L6: Morphological units, word formation processes, lexical semantics, ambiguity1:11:51
- 7Generative AI L7: Word synonymy, similarity, relatedness, one hot vectors, term frequency, tf-idf1:12:37
- 8Generative AI L8: Word embedding concept, training data preparation, prediction based encodings1:00:21
- 9Generative AI L9: Skipgram, CBOW (full softmax variant, negative sampling variant), GloVe1:11:13
- 10Generative AI L10: semantic properties, evaluation and societal biases of word embeddings58:37
- 11Generative AI L11: Methods of mitigating bias, its implications, limitations & related experiments55:29
- 12Generative AI L12: Neural network for next-word-prediction based on 3 context words, intro to RNN29:38
- 13Generative AI L13: RNN, sequence modelling applications, encoder-decoder architecture52:43
- 14Generative AI L14: detailed derivation of BPTT of RNNs, truncated BPTT, teacher forcing55:12
- 15Generative AI L15: Professor forcing, Issues with RNNs, their solutions, intro to LSTM56:45
- 16Generative AI L16: LSTMs35:55
- 17Generative AI L17: BiRNNs and Stacked Deep RNNs17:18
- 18Generative AI L18: Cross-Attention in Encoder-Decoder Sequence Models1:18:15
- 19Generative AI L19: Transformers Intuition1:09:31
- 20Generative AI L20: Transformers in Equations54:44
- 21Generative AI L21: Decoding Strategies Part011:05:57
- 22Generative AI L22: Decoding Strategies Part0241:49
- 23Generative AI L23: Training the transformer51:09
- 24Generative AI L24: Architectural split (encoder only, decoder only and encoder-decoder), BERT1:09:47
- 25Generative AI L25: GPT pre-training objective, span corruption, comparisons of various models51:10
- 26Generative AI L26: LLM jail breaks, DAN jail breaks. Post training and fine-tuning LLMs1:10:33
- 27Generative AI L27: Supervised fine tuning, full fine tuning, parameter-efficient fine tuning (PEFT)58:53
- 28Generative AI L28: prompt tuning, prefix tuning, adapter tuning1:10:59
- 29Generative AI L29: Low Rank Adaptation (LoRA)46:29
- 30Generative AI L30: Quantization55:33
What to take next
Related courses
Videos by Agha Ali Raza - Youtube Channel, played from their official YouTube channel. Subscribe on YouTube
Official upload by the teacher or institution on their own channel, shown with YouTube's embedded player.
Lessons listed from the official YouTube playlist. Full playlist on YouTube
Lessons last checked Oct 2026. Course summary and lesson structure by Sabaq.
Report a problem or a rights request