Metraipolis › archive › dc_505f054302ca
From Production Traffic to Post-Training: Building a Self-Hosted LLM That Covers the Corporate Request Mix — arXiv:2609.01572
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.01572
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-17T06:41:41.004Z
- SHA-256 of the archived text
ad7c4813dfaa2a7d01bb34155beff7dca52e4a01c47c1a560bacfc26750b0159- Stable URL of this page
https://metraipolis.com/documento/dc_505f054302ca
How to cite
From Production Traffic to Post-Training: Building a Self-Hosted LLM That Covers the Corporate Request Mix — arXiv:2609.01572 — arXiv (en), https://arxiv.org/abs/2609.01572 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_505f054302ca (Metraipolis archive dc_505f054302ca, sha256 ad7c4813dfaa2a7d01bb34155beff7dca52e4a01c47c1a560bacfc26750b0159)
BibTeX
@misc{dc505f054302ca,
title = {From Production Traffic to Post-Training: Building a Self-Hosted LLM That Covers the Corporate Request Mix — arXiv:2609.01572},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.01572},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_505f054302ca; sha256 ad7c4813dfaa2a7d01bb34155beff7dca52e4a01c47c1a560bacfc26750b0159},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.01572. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Data-residency constraints force enterprises to self-host LLMs, but continuous adoption of newer models without decommissioning their predecessors expands the serving fleet, fragmenting a finite GPU pool. We consolidate traffic from over 200 internal applications onto a single model by closing quality gaps identified through production error analysis along three axes: instruction following, function-calling, and internal task distribution. Quality is tracked by offline benchmarks stratified to production traffic and scored by deterministic verifiers or calibrated LLM judges. Rather than optimising all objectives jointly, which introduces cross-domain reward interference, we train a separate GRPO expert per axis and merge them via two-stage SLERP. Each expert's reward exposes a distinct failure mode, namely semantic collapse, over-calling, and verbosity hacking, each requiring a domain-specific fix. In non-reasoning mode the recipe surpasses a ${\sim}7\times$ larger by total parameters baseline on the in-house Arena with 69.6 to 65.8, instruction following with 0.85 to 0.83, and function-calling with 0.79 to 0.77, while lifting general dialogue benchmarks. The model absorbs 50% of platform traffic, 116M requests per month, at a fraction of the serving cost.
Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-02 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2609.01572 — 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 2026-10-17T06:33:38.873Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-02?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).