Historical milestone · 2022
Training compute-optimal large language models
Reviewed through September 18, 2026
2022 · Historical milestone
Training compute-optimal large language models
- Era
- 2020s
- Theme
- Machine-learning foundations
- Evidence form
- Controlled scaling study
- School / paradigm
- Scaling laws / empirical optimization
- Institution / context
- Google DeepMind
- Researchers
- Jordan Hoffmann; collaborators
Understand
Plain-language record, transferred from the reviewed source module.
Theory or experimental setup. Varied model size and training-token count under fixed compute budgets and trained a 70-billion-parameter test model on substantially more data.
Result / historical claim. Reported that many large language models were undertrained and that jointly scaling parameters and data improved compute efficiency.
Apply
Professional implication, only where the reviewed record states one.
The checked-in record does not state a separate professional application for this entry. The topic page places it in the wider research lineage: .
Verify
Evidence status, stated limitations, and the external sources this record actually carries.
Evidence form. Controlled scaling study
Limitation / debate. The fitted relationship was empirical, regime-dependent, and did not settle data quality, rights, or downstream reliability.
Source status. This milestone row does not carry a primary-source URL in the approved export, and we do not have a verified link for it in our own research. We do not guess one.
No primary-source URL is recorded for this entry in our reviewed data. Rather than manufacture a citation, we link the Implement Agentic research page that carries the record.
Reproduce
A reproduction tutorial is linked only when one exists for this exact record.
A reproduction tutorial is not yet available for this entry. The closest reviewed material is .
Cite or share
APA-like: This historical record carries a year only, and no author or publisher of record in the checked-in data. An APA reference would have to invent that metadata.
BibTeX: BibTeX requires an author and publication venue. Historical lineage entries store a narrative record and its source link, not structured authorship, so the field would be fabricated.