Deadline-Aware Adaptive Prefill Chunking (arXiv 2609.07883, Wed 9 Sep digest) presents SLOWeave, an online scheduler that selects the largest prefill chunk predicted to finish before the earliest active decode deadline — replacing the fixed chunk size that trades launch overhead against latency spikes. The decision needs no workload-specific tuning and runs as a logarithmic-time search over a monotone iteration-cost model. The authors prove that whenever a decode-only iteration is feasible and the cost predictor is accurate, SLOWeave maximizes immediate prefill progress among deadline-safe choices.
SLOWeave sizes prefill chunks to decode deadlines with a performance guarantee
2026-09-10Frontier watchTRIAL
Deadline-Aware Adaptive Prefill Chunking (arXiv 2609.07883, Wed 9 Sep digest) presents SLOWeave, an online scheduler that selects the largest prefill chunk predicted to finish before the earliest active decode deadline — replacing the fixed chunk size that trades launch overhead against latency spikes. The decision needs no workload-specific tuning and runs as a logarithmic-time search over a monotone iteration-cost model. The authors prove that whenever a decode-only iteration is feasible and the cost predictor is accurate, SLOWeave maximizes immediate prefill progress among deadline-safe choices.
Imported from the existing research archive. Migration preserves the recorded judgment and is not a fresh verification.
What this means for you
Chunked prefill is standard in vLLM/SGLang-class engines, but the chunk size is a hand-tuned constant per deployment. This replaces it with a deadline-driven online decision that provably does not hurt decode SLOs — an afternoon-scale A/B upgrade for serving teams.
FRONTIER WATCH
Public artifacts and validated results are different.
The recorded fact is the public work and its artifacts. Author-reported outcomes do not establish independent reproduction. Next checks include usable code, task conditions and third-party validation.
Evidence & related material
Technical details & scores
| Arxiv | 2609.07883, Wed 9 Sep 2026 digest |
|---|---|
| Method | largest deadline-safe prefill chunk; logarithmic-time search over a monotone cost model; no workload tuning |
| Guarantee | proven max immediate prefill progress among deadline-safe choices when decode-only iteration is feasible and the predictor is accurate |
Technical impact 3
Engineering value 4
Adoption signal 2
Maturity 2
Verification cost 2
Risk 1