LVL 01SK
Project overview
SYSTEM ARCHITECTURE

Transformer from Scratch

Follow the data, decisions, feedback, and validation boundaries before writing the full system.

Transformer from Scratch first-principles architecture infographic

How to read this diagram

Read left to right for the forward path: raw information becomes a representation, passes through the project’s main computational ideas, and produces an output that can be measured. Then follow the feedback path back toward the trainable or decision-making components.

01

Multi-head attention

Attention builds a content-dependent weighted average. Queries describe what each position needs, keys describe what each position offers, and values carry the information. Scaling by the square root of key dimension prevents dot products from pushing softmax into saturation.

Boundary check: document its accepted input, output shape, mutable state, failure modes, and the metric that proves this stage is correct before connecting it downstream.

02

Positional encoding

Attention alone is permutation-equivariant, so it cannot know token order. Positional features inject location into the representation, either as fixed sinusoids, learned vectors, rotations such as RoPE, or relative biases.

Boundary check: document its accepted input, output shape, mutable state, failure modes, and the metric that proves this stage is correct before connecting it downstream.

03

Beam search

Beam search defines one of the project’s main information transformations. Understand its input representation, objective, numerical invariants, computational cost, and failure modes before relying on a library implementation.

Boundary check: document its accepted input, output shape, mutable state, failure modes, and the metric that proves this stage is correct before connecting it downstream.

Architecture review checklist

  • Every arrow has a documented shape, dtype, unit, or schema.
  • Training and evaluation paths cannot leak information into each other.
  • Randomness is seeded and captured in experiment metadata.
  • Expensive stages expose timing, memory, throughput, and error metrics.
  • Each feedback loop has a stop condition and a rollback strategy.
  • Small reference implementations exist for numerical comparisons.