LVL 01SK
Project overview
MATHEMATICS & VISUAL EXPLANATION

Transformer from Scratch

Derive the core equations slowly, attach every symbol to code, and verify the result with small numerical examples before scaling the implementation.

How to study the mathematics

Read each chapter in four passes: intuition, symbols, derivation, and implementation. Recalculate the worked example by hand. Then change one number and predict the direction of the result before running code.

01
FIRST-PRINCIPLES DERIVATION

Multi-head attention

Intuition before notation

Attention is a learned lookup. Each query scores every key, softmax converts scores into a distribution, and that distribution mixes the values.

Loading equation…

Symbol dictionary

Q
queries
K
keys
V
values
M
mask containing 0 or negative infinity
d_k
key dimension

Derive it one move at a time

  1. 1

    Take every query-key dot product.

  2. 2

    Scale by the square root of key width.

  3. 3

    Add padding or causal masks.

  4. 4

    Normalize each query row with stable softmax.

  5. 5

    Use the weights to form a weighted value sum.

INPUTKnown quantities
TRANSFORMMulti-head attention
CHECKValue, shape, invariant

Worked numerical example

For one query with raw scores [2,1], stable softmax subtracts 2 and normalizes [1,e^-1], producing about [0.731,0.269]. The output is 0.731 v1 + 0.269 v2.

Translate the derivation into code

  • Track shapes as batch, heads, query, key, width.
  • Apply masks before softmax.
  • Assert every unmasked attention row sums to one.
02
FIRST-PRINCIPLES DERIVATION

Positional encoding

Intuition before notation

Positional encoding is a transformation whose meaning comes from its domain, codomain, objective, and invariants. The equation is useful only when every symbol maps to a concrete tensor, state, or measurement.

Loading equation…

Symbol dictionary

x,y
input and target
f_theta
parameterized transformation
ell
data objective
Omega
regularizer
lambda
regularization weight

Derive it one move at a time

  1. 1

    Specify the input representation.

  2. 2

    Define the parameterized transformation.

  3. 3

    Choose a loss connected to desired behavior.

  4. 4

    Average over the training evidence.

  5. 5

    Add explicit inductive bias or constraints.

INPUTKnown quantities
TRANSFORMPositional encoding
CHECKValue, shape, invariant

Worked numerical example

For a one-parameter predictor f(x)=theta*x with x=2, y=6, squared loss is (2theta-6)^2. The minimum without regularization is theta=3.

Translate the derivation into code

  • Give every axis a semantic name.
  • Implement a scalar reference first.
  • Compare optimized output and gradients against the reference.
03
FIRST-PRINCIPLES DERIVATION

Beam search

Intuition before notation

Beam search is a transformation whose meaning comes from its domain, codomain, objective, and invariants. The equation is useful only when every symbol maps to a concrete tensor, state, or measurement.

Loading equation…

Symbol dictionary

x,y
input and target
f_theta
parameterized transformation
ell
data objective
Omega
regularizer
lambda
regularization weight

Derive it one move at a time

  1. 1

    Specify the input representation.

  2. 2

    Define the parameterized transformation.

  3. 3

    Choose a loss connected to desired behavior.

  4. 4

    Average over the training evidence.

  5. 5

    Add explicit inductive bias or constraints.

INPUTKnown quantities
TRANSFORMBeam search
CHECKValue, shape, invariant

Worked numerical example

For a one-parameter predictor f(x)=theta*x with x=2, y=6, squared loss is (2theta-6)^2. The minimum without regularization is theta=3.

Translate the derivation into code

  • Give every axis a semantic name.
  • Implement a scalar reference first.
  • Compare optimized output and gradients against the reference.
04
FIRST-PRINCIPLES DERIVATION

Optimization and parameter updates

Intuition before notation

objective and gradient update is a transformation whose meaning comes from its domain, codomain, objective, and invariants. The equation is useful only when every symbol maps to a concrete tensor, state, or measurement.

Loading equation…

Symbol dictionary

x,y
input and target
f_theta
parameterized transformation
ell
data objective
Omega
regularizer
lambda
regularization weight

Derive it one move at a time

  1. 1

    Specify the input representation.

  2. 2

    Define the parameterized transformation.

  3. 3

    Choose a loss connected to desired behavior.

  4. 4

    Average over the training evidence.

  5. 5

    Add explicit inductive bias or constraints.

INPUTKnown quantities
TRANSFORMOptimization and parameter updates
CHECKValue, shape, invariant

Worked numerical example

For a one-parameter predictor f(x)=theta*x with x=2, y=6, squared loss is (2theta-6)^2. The minimum without regularization is theta=3.

Translate the derivation into code

  • Give every axis a semantic name.
  • Implement a scalar reference first.
  • Compare optimized output and gradients against the reference.
05
FIRST-PRINCIPLES DERIVATION

Probability, normalization, and calibration

Intuition before notation

Softmax converts relative logits into positive normalized probabilities while temperature controls how strongly differences are expressed.

Loading equation…

Symbol dictionary

z_i
logit for outcome i
m
maximum logit
tau
temperature
p_i
normalized probability

Derive it one move at a time

  1. 1

    Divide logits by temperature.

  2. 2

    Find the maximum scaled logit.

  3. 3

    Subtract it without changing probability ratios.

  4. 4

    Exponentiate the shifted values.

  5. 5

    Divide by their sum.

INPUTKnown quantities
TRANSFORMProbability, normalization, and calibration
CHECKValue, shape, invariant

Worked numerical example

Logits [1000,999] overflow naively. Subtracting 1000 gives [0,-1], whose probabilities are approximately [0.731,0.269].

Translate the derivation into code

  • Reduce maximum with keepdims.
  • Use the same axis for maximum and sum.
  • Test translation invariance by adding a constant to every logit.
06
FIRST-PRINCIPLES DERIVATION

Evaluation uncertainty and error bars

Intuition before notation

A reported metric is an estimate from finite evidence. Uncertainty separates stable improvement from sampling noise.

Loading equation…

Symbol dictionary

x_i
per-example or per-run measurement
x-bar
sample mean
s
sample standard deviation
n
independent observations

Derive it one move at a time

  1. 1

    Choose the independent unit.

  2. 2

    Compute one measurement per unit.

  3. 3

    Estimate mean and sample variance.

  4. 4

    Convert variation into standard error.

  5. 5

    Report an interval with assumptions or use bootstrap resampling.

INPUTKnown quantities
TRANSFORMEvaluation uncertainty and error bars
CHECKValue, shape, invariant

Worked numerical example

Five seeded scores with mean 0.80 and standard deviation 0.04 have SE about 0.0179, giving a rough 95% interval 0.765 to 0.835.

Translate the derivation into code

  • Store per-example and per-seed values, not only the average.
  • Use stratified or paired intervals when the design requires them.
  • Never treat correlated tokens or timesteps as independent runs.