Stochasticity
The property of potentially producing multiple outputs under the same conditions because a model or system intentionally incorporates probabilistic choices, such as sampling.
Example
Repeatedly generating an answer to the same question under unchanged settings can yield different wording or response candidates due to sampling.
Distinctions and cautions
Stochasticity concerns intentionally introduced probabilistic variation. It is treated separately from nondeterminism arising from execution-environment or numerical differences even when no random sampling is used.
Evidence
View evidence →
LEPO: Latent Reasoning Policy Optimization for Large Language Models
- Title
- LEPO: Latent Reasoning Policy Optimization for Large Language Models
- Author / organization
- Zhou et al.
- Year
- 2026
- Source type / publication status
- Findings ACL peer-reviewed research; published
- Supported concepts / claim context
- Stochasticity as deliberately introduced sampling/randomness.
- Scope and limitations
- A particular latent-reasoning method using stochasticity; not the sole authority for the general definition.
- Access / version note
- Published ACL Findings.
- Recorded official URL / DOI / identifier
- DOI:
10.18653/v1/2026.findings-acl.707(DOI resolver)
CoRun: Padding is Simple and Efficient for Deterministic LLM Inference
- Title
- CoRun: Padding is Simple and Efficient for Deterministic LLM Inference
- Author / organization
- Zhao et al.
- Year
- 2026
- Source type / publication status
- Direct technical preprint; preprint
- Supported concepts / claim context
- Nondeterminism; batch-dependent numerical execution.
- Scope and limitations
- System evidence from a particular serving/hardware stack.
- Access / version note
- August 2026 preprint explicitly included; NOT a published-version substitute.
- Recorded official URL / DOI / identifier
- Preprint reference:
arXiv:2608.14376