RESEARCHMonitorNEXT 12 MONTHS
Greedy Decoding Is Not Precision-Invariant: Cross-Precision Output Divergence in LLM Inference
arXiv cs.LG — Machine Learning
Factual evidence
What the source reports
Research reveals greedy decoding in LLMs is not precision-invariant, with BF16 and FP16 causing output divergence on identical hardware.
Inspect the evidence
- Inclusion basis
- Enterprise AI
- Publisher and source type
- arXiv cs.LG — Machine Learning · RESEARCH
- Published by source
- 23 September 2026
- Collected by OneBench
- 24 Sept 2026, 03:02 UK
Stored source excerpt
arXiv:2609.26621v1 Announce Type: new Abstract: Greedy decoding from large language models is commonly treated as deterministic. We show it is not precision-invariant: the same model,…
Short excerpt from the collected text, not the full source. Use the source link to read it in context.
The factual summary is a OneBench synthesis, not a quotation or independent verification. Collection time is not publication time. Open the source for its full context; related reporting can share the same underlying announcement.