NLP with Transformers

Tokenisation, attention, pretraining and fine-tuning for classification, extraction and sequence tasks with Hugging Face.

⏰ 8 weeks🎯 Standard🧩 Certificate🖥 Any device · our GPUs

Apply for this course Where this leads

Generated from this course

🎯 About this course

NLP has moved fast, but the fundamentals still decide whether a system works. This course teaches them by implementation: train a BPE tokeniser, code masked multi-head attention, align subword labels and decode BIO tags, compare greedy and beam search, and compute the metrics that match real decisions — entity-level F1, macro-F1, ROUGE (and its limits) and calibration error.

You also learn serving: length bucketing, dynamic batching, distillation and confidence-based human review. Graded labs are lightweight NumPy and pure-Python exercises that run offline, and the applied studio has you build a support-email classifier and entity extractor with a model card.

Who it is for

Engineers and data scientists building language features on transformer models.

What makes it different

You implement the mechanisms rather than only calling libraries, so you can debug them.

By the end you will be able to

Explain and implement subword tokenisation and reason about token cost across languages.
Implement masked multi-head self-attention and explain the transformer block, positional information and model families.
Fine-tune encoders for classification and NER, aligning labels to subwords and decoding BIO tags robustly.
Choose and implement decoding strategies (greedy, beam, sampling) for sequence-to-sequence tasks and discuss faithfulness risks.
Evaluate NLP systems with task-appropriate metrics, entity-level scoring, calibration and human evaluation design.
Serve NLP models within latency and cost budgets using batching, bucketing, distillation and fallbacks.
Deliver a classification and extraction service with entity-level evaluation, a review fallback and a model card.

Curriculum

Topics covered

Core topics 6 topics

Tokenisation and embeddings
How text becomes numbers: tokenisation (BPE), vocabularies, and embeddings that capture meaning.
The transformer, in depth
The transformer in depth: self-attention, multi-head, positional information, masking and the encoder / decoder families.
Fine-tuning for classification and NER
Fine-tuning encoders for classification and named-entity recognition, and decoding BIO tags into entities.
Sequence-to-sequence tasks
Sequence-to-sequence tasks: summarisation, translation and generation, decoding strategies and their trade-offs.
Evaluation for NLP
Evaluation for NLP: task-appropriate metrics, macro vs micro averages, ROUGE/BLEU limits, calibration and human evaluation.
Serving and latency
Serving NLP models: dynamic batching, length bucketing, distillation, caching and latency budgets.

Course at a glance

See how this course is built

Where the time goes

Longer spokes mean more hours on that module.

Course roadmap click a step

Try a lesson0 / 3

Sample questions from this course. Your progress is kept only in your browser.

Week by week

Drag the slider, or press play, to walk through the course and see what you will have built by each week.

Week 1 of 8

Wk 1–3

Transformers and fine-tuning

You produceA fine-tuned text classifier

Wk 4–6

Extraction and seq2seq

You produceAn information-extraction or summarisation model

Wk 7–8

Evaluation, serving and project

You produceA served model and an evaluation report

What you'll build

Attention and tokenisation implementations

A BPE trainer and a masked multi-head attention module verified by tests.

Entity-level evaluation report

Classification and NER evaluation with calibration analysis and a threshold recommendation.

Extraction service with model card

A served classification and entity extraction system with evaluation, fallback and a model card.

Why take this course

Understand transformers by implementing them
Evaluate NLP at the level the business uses — entities, not tokens
Labs run offline in seconds, no model downloads
Serving skills: batching, bucketing, distillation, review fallbacks

You will build

A BPE tokeniser trainer
Masked multi-head self-attention in NumPy
A subword label aligner and robust BIO decoder
Greedy and beam-search decoders with length normalisation
ROUGE, macro-F1 and calibration-error evaluators
A dynamic batcher and a classification + extraction service

What you can do afterwards

Debug tokenisation, attention and decoding problems
Choose metrics and thresholds that match the decision
Ship and monitor an NLP service with a model card

Career fit

🧭 Where this fits

Roles whose screened skills map to this course, and the skills it lets you show.

Skills you will build

Evaluation and deployment of the model
Language models — transformers, NER, seq2seq
Tokenisation & text embeddings

How to take it

💳 Delivery & fees

One course, four ways to attend. Fees are per course, in rupees excluding 18% GST (the GST-inclusive amount is shown too) with Canadian-dollar fees alongside.

Your fee includes our GPU lab infrastructure — no cloud credits to buy. See how we compare, with sources →