When Top-1 Fails: Calibrating LoRA Monitors for Masked Diffusion LMs

• URL: http://arxiv.org/abs/2606.24119v1
• Date: Tue, 23 Jun 2026 04:09:58 GMT
• Title: When Top-1 Fails: Calibrating LoRA Monitors for Masked Diffusion LMs
• Abstract summary: Discrete diffusion language model fine-tuning inherits inexpensive diagnostics from confidence monitors, but their PEFT-training meaning is untested.<n>We test top-1 argmax concentration as a collapse warning.<n>Across 816 LoRA/PEFT configurations from three DLM families, the warning fires for every configuration while logs record 0/816 actual collapses at the 200 step horizon, giving zero precision.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Discrete diffusion language model (DLM) fine-tuning inherits inexpensive diagnostics from denoising-time confidence monitors, but their PEFT-training meaning is untested. We test top-1 argmax concentration as a collapse warning. Across 816 LoRA/PEFT configurations from three DLM families, the warning fires for every configuration while logs record 0/816 actual collapses at the 200 step horizon, giving zero precision. The cause is pre-equilibrium saturation: top-1 concentration is already high before optimization and quickly becomes insensitive to final training stability. We then evaluate max LoRA gradient norm, a parameter-side signal that samples gradient routing rather than token concentration. On a pooled held-out LLaDA-family split, a train-optimized threshold identifies top-decile final-loss configurations with precision 0.68 and F1=0.79, above the all-positive top-1 baseline even at the lower split-bootstrap confidence bound. Autoregressive controls and cross-family threshold failures bound the result to short-horizon DLM-LoRA inspection rather than a universal collapse detector. Workflow: drop top-1 as a PEFT alarm, log max-gradient early in training, and calibrate thresholds per DLM family before routing runs for inspection.

Related papers

• One Step, One Lead: Mitigating Higher-Order Interference in Multi-Domain Reinforcement Learning via Cross-Step Control [62.82372407840088]
  We show that same-point domain gradients may remain nearly even when consecutive realized updates partially reverse one another in output space.<n>We propose OSOL, which designates a focus domain at each iteration, to rank token-level rebound risk.<n>OSOL reaches a domain-macro average of 0.4822, improving by 5.7% over the strongest compared baseline, without explicit higher-order differentiation.
  arXiv  Detail & Related papers  (2026-09-06T08:32:59Z)
• YOLO-PEFT: Parameter-Efficient Fine-Tuning on YOLO Family [35.54329849535432]
  YOLO-PEFT is a structure-aware framework that formulates adapter placement as an auditable constraint-planning problem.<n>Within the evaluated detector families, YOLO-PEFT replaces manual target-module trial and error with explicit, inspectable planning.
  arXiv  Detail & Related papers  (2026-08-07T09:59:19Z)
• Onnes: A Physics-Grounded Multi-Agent LLM Simulator for Cryogenic Fault Diagnosis in Quantum Computing Infrastructure [0.0]
  Dilution refrigerators are the enabling infrastructure of superconducting quantum computers.<n>We present Onnes, a physics-grounded digital-twin simulator of a dilution refrigerator.
  arXiv  Detail & Related papers  (2026-07-07T03:57:21Z)
• Uncertainty-Aware Sequential Decision Rules for Event-Triggered LLM Invocation in Streaming Systems [3.556355987197792]
  Streaming inference pipelines increasingly pair lightweight fast models with Large Language Models (LLMs)<n>We cast this as a risk-based sequential stopping problem, where a trigger policy fires when a risk functional over the observation history exceeds a threshold.<n>We prove six results: a minimum inter-event time bound excluding trigger; optimality of threshold policies via smooth pasting; approximate SPRT guarantees under estimated parameters.
  arXiv  Detail & Related papers  (2026-06-27T08:18:09Z)
• SDS-LoRA: Overcoming Anisotropic Gradient Scaling in Low-Rank Adaptation [64.33799467286265]
  Low-Rank Adaptation (LoRA) enables efficient adaptation of large pre-trained models to downstream tasks.<n>We show that when a full fine-tuning gradient is backpropagated to the low-rank matrices, it undergoes anisotropic scaling driven by their singular values.<n>We propose a new low-rank parameterization, SDS-LoRA, which structurally decouples singular values from the backward pass.
  arXiv  Detail & Related papers  (2026-06-15T09:27:50Z)
• Max-Window Scale Estimation for Near-Lossless HiF8 W8A8 Quantization-Aware Training [5.976336341654394]
  Quantization-aware training (QAT) with low-bit floating-point formats enables efficient LLM deployment, yet introduces subtle failure modes invisible to standard training metrics.<n>We present a systematic study of HiF8 W8A8 QAT for OpenPangu-Embedded-1B through the lens of Delayed Scaling (DTS)<n>We identify and disentangle two failure modes: (i)amax saturation, where delayed scale estimates corrupt knowledge-sensitive representations via forward-pass clipping, and (ii)catastrophic forgetting, where an aggressive learning rate overwrites pretrained commonsense knowledge independently of quantization
  arXiv  Detail & Related papers  (2026-05-25T09:19:57Z)
• Revisiting Auxiliary Losses for Conditional Depth Routing: An Empirical Study [31.968379218484746]
  Gate decision must propagate through many layers before it influences the language modeling (LM) loss.<n> Auxiliary losses are commonly stacked to stabilise training, yet the interactions among them -- particularly between a predictive auxiliary and explicit score supervision -- have not been systematically compared under controlled conditions.<n>We trace this to an off-policy oracle label that assumes all subsequent layers execute full, whereas gated execution routes only a fraction through full.
  arXiv  Detail & Related papers  (2026-04-19T03:20:40Z)
• 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)
• Curvature-Guided LoRA: Steering in the pretrained NTK subspace [60.35296431630704]
  We introduce the prediction alignment problem, aiming to match the predictor obtained via PEFT to that of full fine-tuning at the level of outputs.<n>We show that this objective naturally leads to a curvature-aware, second-order formulation, where optimal low-rank updates correspond to a Newton-like, curvature-whitened gradient.<n>Based on this insight, we propose Curvature-Guided LoRA (CG-LoRA), which selects and scales adaptation directions using local curvature information.
  arXiv  Detail & Related papers  (2026-03-31T14:46:39Z)
• REAL: Regression-Aware Reinforcement Learning for LLM-as-a-Judge [83.2858110368572]
  We propose textbfREAL (underlineREgression-underlineAware Reinforcement underlineLThought), a principled RL framework designed to optimize regression rewards.<n>We show that REAL consistently outperforms both regression-aware SFT baselines and standard RL methods.
  arXiv  Detail & Related papers  (2026-03-17T21:19:08Z)
• Information Fidelity in Tool-Using LLM Agents: A Martingale Analysis of the Model Context Protocol [69.11739400975445]
  We introduce the first theoretical framework for analyzing error accumulation in Model Context Protocol (MCP) agents.<n>We show that cumulative distortion exhibits linear growth and high-probability deviations bounded by $O(sqrtT)$.<n>Key findings include: semantic weighting reduces distortion by 80%, and periodic re-grounding approximately every 9 steps suffices for error control.
  arXiv  Detail & Related papers  (2026-02-10T21:08:53Z)
• EDIT: Early Diffusion Inference Termination for dLLMs Based on Dynamics of Training Gradients [6.736735746633275]
  Diffusion-based large language models (dLLMs) refine token generations through iterative denoising, but answers often stabilize before all steps complete.<n>We propose EDIT, an inference-time criterion that adaptively stops denoising once sufficient reasoning stability relative to training-time reasoning is detected.
  arXiv  Detail & Related papers  (2025-11-29T23:47:47Z)
• Scaling with Collapse: Efficient and Predictable Training of LLM Families [8.979516613284174]
  Collapse emerges as a signature of compute-efficient training.<n>We show that collapse emerges as a signature of compute-efficient training.<n>We demonstrate two applications at scale.
  arXiv  Detail & Related papers  (2025-09-29T17:26:11Z)

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.