Week 13 – 04.05 Research on AI – FMP – Critical Practice and Exploration

Stage-by-Stage Terminology Breakdown

Stage	Name	Term (CN/EN)	Brief Description
1️⃣ Input	User Input Text	输入文本 / Input Text	Natural language input, e.g., “I want to write an email”
2️⃣ Tokenization	Tokenizer	分词器 / Tokenizer	Breaks text into “tokens” (words/subwords), e.g., GPT uses BPE
3️⃣ Initial Representation	Embedding	词向量 / Vector Representation	Maps each token to a high-dimensional vector (embedding)
4️⃣ Model Computation	Encoding via Transformer	编码器 / Self-Attention Mechanism	Transformer layers compute contextual representation of tokens
5️⃣ Hidden State Representation	Hidden Representation	中间隐层表征 / Hidden Representation	Intermediate representations of tokens (with context) in each layer
6️⃣ Output Prediction	Output Logits	输出分布 / Probability Representation	Predicts probability distribution for the next token at each position
7️⃣ Decoding	Decoding	采样 / Sampling or Beam Search	Selects the most likely next token (e.g., temperature sampling)
8️⃣ Output	Model Text Output	模型文本结果 / Model Output	Final human-readable output, e.g., “Hello, here is your email draft…”