Train a causal language model
Use train a causal language model to move the transformer from scratch production brief toward a defensible release.
Domain autocomplete needs a model whose every component the team can explain.
This lesson isolates train a causal language model as one decision inside that system. The people affected are product users and operators; the learning data must carry event time, availability time, ownership, and version; and the operating envelope is one-device educational build.
Choose whether and how to use multi-head attention at a declared prediction cutoff.
Measure decision utility alongside calibration, slice reliability, and system latency—not model score alone.
Domain autocomplete needs a model whose every component the team can explain. An unsafe release must degrade to a named baseline or the last known-good version.
Build the mental model before the machinery.
The core move is to treat train a causal language model as a contract between data, a computation, and an action. Version tokenizer, vocabulary, special tokens, architecture, weights, prompt template, and generation policy together. The implementation becomes easier to debug once you can state which inputs exist, which state is learned, what output means, and what must remain invariant after serialization.
multi-head attention
Define it in a hand-checkable form and name the prediction-time inputs.
pre-norm
Connect it to the production metric and identify what it cannot guarantee.
next-token loss
Stress it with a slice, a temporal boundary, and a failure-safe alternative.
KV cache
Stress it with a slice, a temporal boundary, and a failure-safe alternative.
Lessons 1, 2, 3 in this course.
Name every symbol. Check every shape.
Scaled dot-product attention is the central invariant for this lesson. The formula is useful only when its inputs match the production cutoff and its output maps to an action.
Scaled dot-product attention
| Symbol | Meaning / shape / unit |
|---|---|
Q, K, V | query, key, value matrices |
d_k | head width |
M | causal or padding mask |
Open derivation and numerical substitution
Start from the production quantity being optimized, substitute the observed values with their declared units, then isolate the model-controlled term. Preserve shape annotations at each step so broadcasting or aggregation cannot silently change the result.
- Write the named inputs: Q, K, V, d_k, M.
- Substitute one small, hand-checkable batch before vectorizing.
- Calculate an independent reference value and compare within a declared tolerance.
# equation → code contract
inputs = validate_shapes_and_units(batch)
value = compute_c14(inputs)
assert is_finite(value)Calculate it small. Shape it realistically. Break it on purpose.
A result you can reproduce on paper
Compute one attention head for four tokens and alter one key to watch a softmax row move.
- Write every input and unit.
- Substitute values into the scaled dot-product attention equation above.
- Compare the result to one simple baseline and explain the direction of the difference.
The same reasoning under real constraints
Train a small decoder on versioned support text and serve bounded, KV-cached completions.
The production record includes the data snapshot, transformation state, artifact identity, cutoff, score, decision, and the version of the policy that consumed it.
The attractive result you should reject
Without the causal mask, each position sees its target and validation loss becomes impossibly good.
Diagnostic: replay the smallest failing slice from immutable inputs, then compare each boundary rather than retuning the model.
Change one assumption and make the tradeoff visible.
This lab runs predefined TypeScript only. It never executes learner code. Use the slider, numeric input, reset, live text, or table—the computation is the same.
Attention temperature lab
Change score temperature and inspect the exact softmax distribution.
Assumption: Fixed scores [1.2, 0.3, −0.4, 1.8].
Open nonvisual data table
| Item | Computed state | Interpretation |
|---|---|---|
| token 1 | 1.2 | 0.2915 |
| token 2 | 0.3 | 0.1185 |
| token 3 | -0.4 | 0.0589 |
| token 4 | 1.8 | 0.5311 |
Trace the complete operating path.
- 01
Validate and version multi-head attention.
- 02
Compute train a causal language model from prediction-time-safe inputs.
- 03
Persist model, feature, and configuration identities together.
- 04
Serve or materialize behind explicit one-device educational build.
- 05
Join telemetry to mature outcomes and retain a rollback path.
Observability
Join service health, input quality, prediction distributions, slice behavior, and mature outcomes by exact version.
Cost
Measure storage, preprocessing, compute, queueing, and human review under a representative arrival pattern.
Failure modes
Without the causal mask, each position sees its target and validation loss becomes impossibly good. Add a detector, owner, mitigation, and stop condition for this class of failure.
Alternatives
Compare a rule, a simpler statistical baseline, and a different system boundary before adding model complexity.
Explain the contract, not just the vocabulary.
Decoder-only transformer from scratch
Specify tokenizer, attention blocks, causal training, unit tests, sampling, and KV-cache parity without a transformer module.
- Shape/mask tests
- Tiny-batch overfit
- Held-out evaluation
- Cached/uncached parity
Read primary material with a purpose.
Return to the opening failure.
Version tokenizer, vocabulary, special tokens, architecture, weights, prompt template, and generation policy together.
For this lesson, the release evidence is a hand-checked formal result, deterministic simulation output, a ≥80% checkpoint, the production rubric, and a named fallback. The course resolves when the system can produce tokenizer, attention, rope, residual stack, training loop, generation, and kv cache.