teacher forcing concept
| Definition | Feeding a language model the correct text as context rather than its own earlier outputs. Chapter 0 section 0.11. Also teacher-forced. |
|---|---|
| Example | At step 5 the model is fed the reference’s fifth token, not its own fourth guess. |
| Book | Data Mining as Observation, draft 0.2, commit f3914f0; entry id teacher-forcing, kind concept. |
| Status | no ledger row names this entry. Corrections: none recorded. |
| Defining equation | none |
| Assumptions and scope |
|
| Prior art | none recorded |
| Evidence | lean/DataMiningAsObservation/TeacherForcing.lean |
| Reviewed | not yet reviewed; generated 2026-09-10 from records at the commits on the provenance page. |
Equation
none
Conditions
- Feeding a model the correct text as context rather than its own earlier outputs. The forced and free runs agree exactly as long as the forced predictions match the reference, and they part at the first token the model gets wrong, after which the free run’s context is no longer the reference’s.
- Perplexity is measured under teacher forcing, so a good perplexity does not certify a free run. The C-12 record’s thirteen-point difference vanished under teacher forcing, and the operator-drift claim it was meant to support is refuted in the ledger.
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 program’s C-12 record in readscope,
readscope/CALIBRATION.md:600-660.
Measurements
none
Failures and corrections
none
Invariance envelope
none declared
Machine checked
lean/DataMiningAsObservation/TeacherForcing.lean,
theorems free_eq_take, free_length,
free_ne_of_error, at observation-data-mining f3914f0; what
the check covers is stated in the book’s appendix
C.
Used in
Data Mining as Observation chapters 0, 11.
Related
perplexity; drift; harness; deployment mismatch.
See also
Book equations stated beside the entry’s terms, not defining it: 0.22, 11.1.
Ledger rows that cite the entry’s records without naming it: OT-4.
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.