Series

NLP & LLM Foundations

A practical research-oriented path through tokenization, embeddings, seq2seq, attention, Transformers, fine-tuning, PEFT, prompting, and NLP/LLM evaluation.

Start with tokenization
中文
NLP posts

Part I

Language as Model Input

  1. 1. Tokenization and Subwords
  2. 2. Word Embeddings and Sentence Embeddings

Part II

Architectures

  1. 1. Sequence-to-Sequence Models
  2. 2. Attention Mechanism
  3. 3. Transformers for NLP

Part III

Adaptation & Interaction

  1. 1. Fine-Tuning Transformers
  2. 2. Parameter-Efficient Fine-Tuning
  3. 3. Prompting and Instruction Following

Part IV

Evaluation

  1. 1. NLP Evaluation
  2. 2. LLM Evaluation and Failure Modes

How this series is different

It is not a generic NLP tutorial. It connects every foundation topic to multilingual AI, domain MT, LoRA/PEFT, RAG, and evaluation.

Concept

Jurafsky & Martin, CS224N, and the Illustrated Transformer provide the conceptual backbone.

Practice

Hugging Face Course, Transformers, Tokenizers, Evaluate, and PEFT provide implementation patterns.

Research

MT, multilinguality, LoRA, COMET, and LLM failure modes connect the basics to my research profile.

Start where language becomes model input.

Tokenization is the quiet first decision that shapes everything a multilingual model can learn.

Read the first note
中文版本