arXiv AI
LLM Post-Training as Brownfield Maintenance: An Industrial Perspective on Dataware Engineering
Read the original on arXiv AI →The Flow has not summarised this story yet — read it at arXiv AI.
Related stories
arXiv AI
Jul 21
CADENCE: Closing the Reasoning Gap via Coverage-Adaptive On-Policy Distillation
arXiv:2607. 16955v1 Announce Type: cross Abstract: On-policy knowledge distillation transfers reasoning from large teachers to compact students, but existing approaches suffer three compounding failure modes: (i) cold-start collapse, where a fresh student assigns near-zero mass to teacher-preferred tokens; (ii) state-agnostic divergence scheduling, where time-only forward/reverse-KL interpolation ignores the student's coverage state; and (iii) binary reward sparsity, where pass/fail signals discard information from partially correct traces.
arXiv AI
Aug 26
PROOF-Gen: From Optimized Data to Better Distillation
PROOF-Gen is a method that improves distillation of tool‑calling models by recovering successful trajectories from teacher failures. It uses per‑scenario prompt optimization to generate corrective guidance that steers the teacher to a passing trajectory, then removes this guidance before training so the student learns from clean demonstrations. On τ2‑bench, PROOF-Gen recovers 93% of failed scenarios, boosting Qwen3‑4B‑Instruct‑2507’s Pass^1 from 0.132 to 0.529 and improving Gemma 4 E4B‑it by 7.2pp on BFCL v4 multi‑turn, while also raising deployed on‑device model performance by up to 5.0pp across response‑quality metrics.
arXiv Machine Learning
Aug 21
When to Retrain: An Empirical Study of Retraining Policies for Streaming ML Under Concept Drift, Budget, and Latency Constraints
arXiv:2608. 19488v1 Announce Type: new Abstract: Production machine learning systems degrade under concept drift, yet practitioners have little principled guidance on when to retrain.
arXiv AI
Aug 24
UpgradeBench: A Decision-Centric Benchmark for Upgrading Fine-Tuned LLM Specialists
UpgradeBench is a decision‑centric longitudinal benchmark that evaluates how fine‑tuned language‑model specialists should be handled when new base‑model releases occur. It covers four consecutive Qwen releases, a continuation checkpoint, six tasks, two model sizes, and OLMo checkpoints with known training lineage, and examines whether retraining, adapter transfer, or other recovery strategies improve specialist performance. The benchmark reveals that upgrade gains vary by task and release interval, that direct adapter copying is sensitive to pretraining distance, and that teacher relabeling can recover specialists without new annotations. "whyItMatters":"The study provides actionable insights into the cost‑effective management of specialist models across model releases, showing how to balance retraining effort with performance gains."
arXiv AI
Aug 20
Open-MOPD: Diagnosing and Fixing Capability Imbalance in Multi-Teacher On-Policy Distillation
Open-MOPD addresses the capability imbalance problem in multi-teacher on-policy distillation (M-OPD) by isolating capability integration from routing ambiguity and revealing a 35.6% headroom gap compared to a domain-routed oracle ensemble. The study identifies three key factors—sequence-length disparities, convergence drift, and reward staleness—that misallocate token-level optimization budgets, leading to severe degradation in concise tasks. The proposed Open-MOPD framework introduces token-share balancing, gap-aware dynamic budget allocation, and student reward refresh, boosting headroom recovery to 83.4% and providing an open-source, reproducible post‑training recipe and evaluation suite.
arXiv AI
Sep 10
Feyospace-v1: How the Cyber Mercury Seven Trained Frontier Cyber Models
Feyospace‑v1 describes a data‑centric framework that enables a seven‑person team to train open‑weight cyber agents, addressing bottlenecks such as environment cost, supervision, and teacher access. The system combines five complementary components—Choulea, SkyReal, Hongzwang, PSBreakup, and Kreator—to analyze reasoning, reduce sampling costs, bypass API limits, restore merged model capabilities, and convert expert interventions into trainable signals. Using a diverse data engine, the team produced 164,269 verified trajectories for supervised fine‑tuning, achieving an average 23.76% improvement on CyberGym and 10.49% on pooled CTF suites, with the final checkpoint ranking 10th on the CyberGym leaderboard and first among comparable‑scale models.