The Observation Theory EncyclopediaFrom TSKAboutBy kindBy chapterBy Lean fileLedgerProvenance

language model concept

DefinitionA model that reads a sequence of tokens and outputs a probability for the next one, scored by perplexity. Chapter 0 section 0.11.
ExampleA model with perplexity 12 on a text is as uncertain, per token, as a fair choice among 12 tokens.
BookData Mining as Observation, draft 0.2, commit f3914f0; entry id language-model, kind concept.
Statusno ledger row names this entry. Corrections: none recorded.
Defining equation

Book equation 0.22.

Assumptions and scope
  • A model that reads a sequence of tokens and outputs a probability for the next one. Its perplexity is two to the power of the average bits per token, at least one and equal to the vocabulary size for a uniform guess, and its attention weights are a softmax, positive and summing to one.
  • A seven-billion-parameter model served passage retrieval over contexts of fourteen thousand tokens across three registered attempts, and a key quantizer with reconstruction cosine 0.995 broke it to a perplexity near ten thousand.
Prior artnone recorded
Evidencelean/DataMiningAsObservation/Perplexity.lean, lean/DataMiningAsObservation/Softmax.lean
Reviewednot yet reviewed; generated 2026-09-10 from records at the commits on the provenance page.
querykey 1key 2key 3key 4key 5key 6key 7key 8softmax weights, sum to one
A model that reads tokens and outputs a probability for the next one.

Equation

Book equation 0.22.

\[\mathrm{PPL}=2^{H},\qquad H=-\frac1T\sum_{t=1}^{T}\log_2 p\big(w_t\mid w_{<t}\big).\]

Conditions

Conditions are curated in entries.toml rather than read from a record.

Ledger

none

First stated

Chapter 0 section 0.11 of Data Mining as Observation, with the serving measurements in geometric-observation/experiments/GO-kv-serving-flip-NOTES.md.

Measurements

none

Failures and corrections

none

Invariance envelope

none declared

Machine checked

lean/DataMiningAsObservation/Perplexity.lean, theorems bits_nonneg, one_le_perplexity, perplexity_uniform, perplexity_mono, bits_of_finding, at observation-data-mining f3914f0; what the check covers is stated in the book’s appendix C.

lean/DataMiningAsObservation/Softmax.lean, theorems denom_pos, softmax_pos, softmax_sum, softmax_shift, softmax_lt_iff, at observation-data-mining f3914f0; what the check covers is stated in the book’s appendix C.

Used in

Data Mining as Observation primer S, chapters 0, 1, 4, 6, 8, 11, 12, 13.

Related

token; perplexity; attention; KV cache; softmax.

See also

Book equations stated beside the entry’s terms, not defining it: 0.23, 11.1.

Ledger rows that cite the entry’s records without naming it: NEG-16 (KV serving, end-task), NEG-2.

Sources-table rows that share a record with the entry without naming it: chapter 1 section 1.4, chapter 2 section 2.5, chapter 8 section 8.2, chapter 11 section 11.2, chapter 13 section 13.6.

Status

Generated 2026-09-10 by encyclopedia/generate.py; book at observation-data-mining f3914f0; the commit of every record is listed in the encyclopedia’s provenance.

← Landauer's principleLaplacian →