Procedural-skill SFT across capacity tiers: A W-Shaped pre-SFT Trajectory and Regime-Asymmetric Mechanism on 0.8B-4B Qwen3.5 Models

• URL: http://arxiv.org/abs/2605.11907v2
• Date: Wed, 13 May 2026 22:47:52 GMT
• Title: Procedural-skill SFT across capacity tiers: A W-Shaped pre-SFT Trajectory and Regime-Asymmetric Mechanism on 0.8B-4B Qwen3.5 Models
• Abstract summary: We measure procedural-skill SFT contribution across three Qwen3.5 dense scales.<n>The SFT-attributable procedural-$$ lift is roughly uniform across sizes.
• Score: 0.0
• License: http://creativecommons.org/publicdomain/zero/1.0/
• Abstract: We measure procedural-skill SFT contribution across three Qwen3.5 dense scales (0.8B, 2B, 4B) on a 200-task / 40-skill holdout, with Claude Haiku 4.5 as a frontier reference. The corpus is 353 rows of (task + procedural-skill block, Opus chain-of-thought, judge-pass) demonstrations. Main finding. Under matched-path LLM-only scoring, the SFT-attributable procedural-$Δ$ lift is roughly uniform across sizes: $+0.070 / +0.040 / +0.075$ at 0.8B / 2B / 4B. Variation in post-SFT $Δ$ ($-0.005$, $+0.100$, $+0.065$) is dominated by a W-shaped pre-SFT base trajectory ($-0.075$, $+0.060$, $-0.010$, Haiku-4-5 at $+0.030$): the 5-step procedure hurts 0.8B and 4B, helps 2B, and helps frontier Haiku modestly. SFT works hardest in absolute terms where the base struggles with the procedure -- a regime-asymmetric pattern with a falsifiable prediction at 8B/14B. Methodology. (i) A bench format-compliance artifact: 83.5% of the holdout uses a deterministic ANSWER-line extractor that under-counts free-form-prose conclusions; our LLM-only re-judge reveals it was systematically biased against the curated condition. (ii) A negative-iteration sequence at 0.8B: three well-formed recipe variants cluster post-SFT curated pass-rate within a 2 pp band, constraining the absolute-pass-rate ceiling to base capacity rather than recipe. Cross-family judge validation. GPT-5.4 via OpenRouter on all 7 configurations (2800 paired episodes) agrees on the direction of every per-student finding: Cohen's $κ\geq 0.754$, agreement $\geq 93.25\%$, max headline $Δ$ shift $\leq 0.035$ pp. Two earlier framings -- "format-only learning at 0.8B" and "SFT contribution shrinks at 4B" -- were path-mismatch artifacts; this paper supersedes both. Single-seed evaluation; threats itemised in the paper.

Related papers

• Online Non-Monotone DR-Submodular Maximization Matching the Offline $0.401$ Factor [47.30196146041476]
  We study asymmetric non-negative, non-monotone DR-submodular functions over compact convex subsets of the $d$-oracle unit.<n>Online algorithm attains factor 0.401$ with sub approximate regret when feedback is conditionally unbiased and bounded.
  arXiv  Detail & Related papers  (2026-09-02T06:00:10Z)
• Adversarial Resilience of Poisson-Process Submodular Maximization over Matroids: From Robust Offline Optimization to Full-Bandit Learning [55.29259818039367]
  We study the nonnegative submodular subject to a general matroid when the offline algorithm is given an arbitrary controlled value oracle.<n>Our algorithm retains limiting approximation factors $1/e$ for non-monotone objectives and $1-1/e$ for monotone objectives.<n>As a consequence, the offline-to-online reduction yields single-bandit CMAB algorithms for general matroid-constrained submodular rewards.
  arXiv  Detail & Related papers  (2026-08-12T14:54:15Z)
• Round-Trip Consistency: Bidirectional Diffusion Models Can Predict Their Own Rollout Errors [51.56484100374058]
  Autoregressive models accumulate error over long rollouts, yet at deployment there is no ground truth to measure it against.<n>We train a single conditional latent diffusion model that steps a dynamical system forward or backward in time via a direction flag.<n>We show that this bidirectionality supplies a measurement-free test-time error signal.
  arXiv  Detail & Related papers  (2026-08-01T13:49:46Z)
• Quantifying Diversity of Thought: A Predictive Law of Weighted LLM Ensemble Lift [0.0]
  This paper provides an experimentally verified formal law for calculating the uplift that diversity of thought provides in Large Language Model ensembles.<n>We test the law on 767,520 inferences from ten open-weight models over two graduate-level science benchmarks.
  arXiv  Detail & Related papers  (2026-07-19T19:01:06Z)
• Imbuing Large Language Models with Bidirectional Logic for Robust Chain Repair [44.80087038178069]
  We introduce Teleological Reasoning Infilling (TRI), a training framework that endows decoder-only transformers with a native emphgoal-conditioned bridging capability.<n>At inference, TRI operates as a surgical repair module within a dual-system loop.<n>Experiments on three benchmarks demonstrate that TRI achieves state-of-the-art performance across all tasks, while reducing per-problem token expenditure by 31.2%.
  arXiv  Detail & Related papers  (2026-06-03T15:58:48Z)
• When Quantization Is Free: An int4 KV Cache That Outruns fp16 on Apple Silicon [0.0]
  KV-cache quantization is framed as a quality--latency trade-off.<n>We show it is emphinverted on Apple Silicon's unified memory.
  arXiv  Detail & Related papers  (2026-05-07T05:44:39Z)
• Scaling Federated Linear Contextual Bandits via Sketching [49.12000877146222]
  This paper proposes Federated Sketch Contextual Linear Bandits (FSCLB)<n>Experiments on both synthetic and real-world datasets show that FSCLB significantly reduces computational and communication costs by over 90 %.
  arXiv  Detail & Related papers  (2026-05-01T08:22:06Z)
• Borrowed Geometry: Computational Reuse of Frozen Text-Pretrained Transformer Weights Across Modalities [0.0]
  Gemma 4 31B weights pretrained exclusively on text tokens, unmodified, transfer across modality boundaries.<n>Gemma 4 31B is the only model on the small-scale frontier as of April 2026.
  arXiv  Detail & Related papers  (2026-05-01T01:23:37Z)
• SHIFT: Robust Double Machine Learning for Average Dose-Response Functions under Heavy-Tailed Contamination [0.0]
  SHIFT (Self-calibrated Heavy-tail Inlier-Fit with Tempering) is a robust DML estimator.<n>It combines cross-fit nuisanceization with a kernel-local Welsch-loss second stage optimized by Graduated Non-Convexity.<n>It has competitive worst-case shape recovery across 1,400 mainsweep fits.
  arXiv  Detail & Related papers  (2026-04-30T19:51:35Z)
• SOAR: Self-Correction for Optimal Alignment and Refinement in Diffusion Models [48.335262141752715]
  Post-training pipeline for diffusion models has two stages: supervised fine-tuning (SFT) on curated data and reinforcement learning (RL) with reward models.<n>We propose SOAR (Self-Correction for Optimal Alignment and Refinement), a bias-correction post-training method that fills this gap.<n>It is on-policy, reward-free, and provides dense per-time supervision with no credit-assignment problem.
  arXiv  Detail & Related papers  (2026-04-14T11:45:15Z)
• Calibration Collapse Under Sycophancy Fine-Tuning: How Reward Hacking Breaks Uncertainty Quantification in LLMs [0.0]
  We investigate whether sycophantic reward signals degrade calibration.<n>Post-hoc matrix scaling applied to all three models reduces ECE by $40$--$64%$ and improves accuracy by $1.5$--$3.0$ percentage points.<n>These findings establish a methodology for evaluating the calibration impact of reward hacking and motivate calibration-aware training objectives.
  arXiv  Detail & Related papers  (2026-04-12T11:09:49Z)
• Topological Characterization of Churn Flow and Unsupervised Correction to the Wu Flow-Regime Map in Small-Diameter Vertical Pipes [0.0]
  We introduce the first topology-based characterization using Characteristic Surfaces (ECS)<n>This work provides the first mathematical definition of churn flow and demonstrates that unsupervised topological descriptors can challenge and correct mechanistic models.
  arXiv  Detail & Related papers  (2026-04-07T17:59:15Z)
• Humains-Junior: A 3.8B Language Model Achieving GPT-4o-Level Factual Accuracy by Directed Exoskeleton Reasoning [0.0]
  Humans-Junior is a 3.8B model that matches GPT-4o on the FACTS Grounding public subset within a $pm 5$ pp equivalence margin.<n>Our approach combines minimal directed "Exoskeleton Reasoning" scaffolds with behavioral fine-tuning that teaches protocol compliance.
  arXiv  Detail & Related papers  (2025-10-29T20:12:36Z)
• Numerical Fragility in Transformers: A Layer-wise Theory for Explaining, Forecasting, and Mitigating Instability [0.0]
  We give a first-order, module-wise theory that predicts when and where errors grow.<n>For self-attention we derive a per-layer bound that factorizes into three interpretable diagnostics.<n>We also introduce a precision- and width-aware LayerNorm indicator $rho_rm LN$ with a matching first-order bound.
  arXiv  Detail & Related papers  (2025-10-17T01:03:02Z)
• On the $O(\frac{\sqrt{d}}{T^{1/4}})$ Convergence Rate of RMSProp and Its Momentum Extension Measured by $\ell_1$ Norm [54.28350823319057]
  This paper considers the RMSProp and its momentum extension and establishes the convergence rate of $frac1Tsum_k=1T.<n>Our convergence rate matches the lower bound with respect to all the coefficients except the dimension $d$.<n>Our convergence rate can be considered to be analogous to the $frac1Tsum_k=1T.
  arXiv  Detail & Related papers  (2024-02-01T07:21:32Z)
• Contextual Combinatorial Bandits with Probabilistically Triggered Arms [55.9237004478033]
  We study contextual bandits with probabilistically triggered arms (C$2$MAB-T) under a variety of smoothness conditions. Under the triggering modulated (TPM) condition, we devise the C$2$-UC-T algorithm and derive a regret bound $tildeO(dsqrtT)$.
  arXiv  Detail & Related papers  (2023-03-30T02:51:00Z)
• Revisiting Weighted Strategy for Non-stationary Parametric Bandits [82.1942459195896]
  This paper revisits the weighted strategy for non-stationary parametric bandits. We propose a refined analysis framework, which produces a simpler weight-based algorithm. Our new framework can be used to improve regret bounds of other parametric bandits.
  arXiv  Detail & Related papers  (2023-03-05T15:11:14Z)
• On Submodular Contextual Bandits [92.45432756301231]
  We consider the problem of contextual bandits where actions are subsets of a ground set and mean rewards are modeled by an unknown monotone submodular function. We show that our algorithm efficiently randomizes around local optima of estimated functions according to the Inverse Gap Weighting strategy.
  arXiv  Detail & Related papers  (2021-12-03T21:42:33Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.