Metraipolis › archive › dc_789c445eb39e

Reproducible LoRA Inference Through Deterministic Split-K Reduction — arXiv:2611.10334

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.10334
License (as archived)
CC0-1.0
Archived
2026-11-10T18:42:23.362Z
SHA-256 of the archived text
e023f0557d817f585801563269358ec890102099986bb5a99756c819890f9b50
Stable URL of this page
https://metraipolis.com/documento/dc_789c445eb39e

How to cite

Reproducible LoRA Inference Through Deterministic Split-K Reduction — arXiv:2611.10334 — arXiv (en), https://arxiv.org/abs/2611.10334 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_789c445eb39e (Metraipolis archive dc_789c445eb39e, sha256 e023f0557d817f585801563269358ec890102099986bb5a99756c819890f9b50)
BibTeX
@misc{dc789c445eb39e,
  title = {Reproducible LoRA Inference Through Deterministic Split-K Reduction — arXiv:2611.10334},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.10334},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_789c445eb39e; sha256 e023f0557d817f585801563269358ec890102099986bb5a99756c819890f9b50},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.10334. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Reproducibility is important for LoRA-adapted language models used in repeated evaluation, auditing, and decision-making. However, Split-K kernels used to introduce parallelism in low-rank LoRA multiplications can have the unintended consequence of numerical nondeterminism when partial results are combined using floating-point atomic additions. We investigate this non-determinism in vLLM across GSM8K mathematical reasoning, MBPP code generation, and Jev-style structured toxicity decisions on Civil Comments and introduce a deterministic Triton Split-K implementation. Under identical-input replay, Atomic Split-K=8 produces different FP32 shrink outputs in 97.61% of comparisons, while our Deterministic Split-K=8 shows no observed variation. Repeated atomic inference changes generated outputs in 17.30% of GSM8K and 14.96% of MBPP comparisons, sometimes altering task correctness. Structured-decision probabilities vary in 81.47% of atomic comparisons, with decision flips. Controlled tracing demonstrates how shrink-output differences propagate into the KV cache and how accumulated cache differences can change token decisions and final answers. Our deterministic Triton Split-K=8 implementation preserves parallelism, achieving 16.8-21.1% latency-based speedup over unsplit reduction with only 1.02-2.62% additional latency relative to Atomic Split-K=8.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI); Distributed, Parallel, and Cluster Computing (cs.DC); Performance (cs.PF).
Pagina del lavoro: https://arxiv.org/abs/2611.10334 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-11-10T18:37:37.583Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).