OPD updates occupy a relaxed off-principal regime and rapidly lock into a low-dimensional subspace that is functionally sufficient for its performance, distinct from SFT and RLVR trajectories.
hub
Pan, Zhangyang Wang, Yuandong Tian, and Kai Sheng Tai
18 Pith papers cite this work. Polarity classification is still indexing.
18
Pith papers citing it
hub tools
citation-role summary
background 1
citation-polarity summary
years
2026 18roles
background 1polarities
background 1representative citing papers
Three problem-level trajectory features derived from the distributional signature of failed LLM rollouts enable failure clustering at 84.3% accuracy and a training-free routing rule that improves rescue by 12.2% on hard cases.
PEFT-Arena reveals distinct stability-plasticity profiles across PEFT methods, with orthogonal finetuning achieving the best Pareto frontier under comparable parameter budgets, supported by weight-space spectral and activation-space retention analyses.
Pion modifies Muon's Newton-Schulz iterations into a controllable high-pass filter that anchors dominant singular values at 1 while suppressing noisy tails, outperforming Muon and AdamW in VLA and RLVR regimes.
On-policy distillation produces coordinate-sparse, FFN-heavy updates that are full-rank but spectrally concentrated away from principal singular subspaces and near-zero source weights.
RELEX extrapolates LLM checkpoints from short RLVR prefixes by projecting deltas onto a rank-1 subspace and fitting a linear trend, matching full training performance at 15% of the steps.
FuRA uses block tensor-train factorization with fixed pretrained SVD basis to achieve full-rank spectral preconditioning, outperforming Full FT by +1.37 on LLaMA-3-8B commonsense reasoning and surpassing QLoRA in quantized settings.
RPSFT improves the in-domain versus out-of-domain performance trade-off during LLM supervised fine-tuning by penalizing rotations in pretrained singular subspaces as a proxy for loss-sensitive directions.
DoTS decouples SFT and RLVR training then synthesizes their task vectors at inference time to match integrated training results at ~3% compute cost.
PULSE exploits BF16-invisible sparsity in weight updates to enable over 100x lower communication in distributed RL post-training via compute-visible sparsification.
TGR performs manifold-informed latent foresight search to boost trajectory coverage in long-context reasoning tasks by up to 13 AUC points with minimal overhead.
Orthonormal initialization for LoRA in RLVR achieves the minimal gap to full fine-tuning, stabilizes training, and outperforms standard LoRA and prior variants on mathematical reasoning benchmarks.
MAST ranks attention-projection tensors by off-principal energy, update magnitude, and forget-gradient coupling to selectively unlearn RLVR-induced reasoning, achieving significant forgetting on MATH while preserving GSM8K and retain MATH unlike full-parameter updates.
RLVR induces sparse off-principal updates forming near-orthogonal shortcuts that degrade merging, addressed via Sensitivity-aware Resolving Merging using Fisher sensitivity, sparsification, and rescaling.
PEFT adapters are positioned as persistent personal state on foundation models, organized via Scale Up, Scale Down, and Scale Out axes, with MinT as an infrastructure example for managing them.
Pion is an optimizer that preserves the singular values of weight matrices in LLM training by applying orthogonal equivalence transformations.
On-policy distillation gains efficiency from early foresight in module allocation and update directions, which the proposed EffOPD method exploits for 3x faster training with comparable performance.
SparseRL-Sync achieves lossless weight synchronization in large-scale RL by sending only changed parameters, reducing communication volume by roughly 100x under observed 99%+ element-level sparsity.
citing papers explorer
-
On the Geometry of On-Policy Distillation
OPD updates occupy a relaxed off-principal regime and rapidly lock into a low-dimensional subspace that is functionally sufficient for its performance, distinct from SFT and RLVR trajectories.
-
Failed Reasoning Traces Tell You What Is Fixable (But Not by Reading Them)
Three problem-level trajectory features derived from the distributional signature of failed LLM rollouts enable failure clustering at 84.3% accuracy and a training-free routing rule that improves rescue by 12.2% on hard cases.
-
PEFT-Arena: Understanding Parameter-Efficient Finetuning from a Stability-Plasticity Perspective
PEFT-Arena reveals distinct stability-plasticity profiles across PEFT methods, with orthogonal finetuning achieving the best Pareto frontier under comparable parameter budgets, supported by weight-space spectral and activation-space retention analyses.
-
Dense Supervision, Sparse Updates: On the Sparsity and Geometry of On-Policy Distillation
On-policy distillation produces coordinate-sparse, FFN-heavy updates that are full-rank but spectrally concentrated away from principal singular subspaces and near-zero source weights.
-
You Only Need Minimal RLVR Training: Extrapolating LLMs via Rank-1 Trajectories
RELEX extrapolates LLM checkpoints from short RLVR prefixes by projecting deltas onto a rank-1 subspace and fitting a linear trend, matching full training performance at 15% of the steps.
-
FuRA: Full-Rank Parameter-Efficient Fine-Tuning with Spectral Preconditioning
FuRA uses block tensor-train factorization with fixed pretrained SVD basis to achieve full-rank spectral preconditioning, outperforming Full FT by +1.37 on LLaMA-3-8B commonsense reasoning and surpassing QLoRA in quantized settings.
-
Rotation-Preserving Supervised Fine-Tuning
RPSFT improves the in-domain versus out-of-domain performance trade-off during LLM supervised fine-tuning by penalizing rotations in pretrained singular subspaces as a proxy for loss-sensitive directions.
-
Understanding and Exploiting Weight Update Sparsity for Communication-Efficient Distributed RL
PULSE exploits BF16-invisible sparsity in weight updates to enable over 100x lower communication in distributed RL post-training via compute-visible sparsification.
-
The Geometric Reasoner: Manifold-Informed Latent Foresight Search for Long-Context Reasoning
TGR performs manifold-informed latent foresight search to boost trajectory coverage in long-context reasoning tasks by up to 13 AUC points with minimal overhead.
-
Geometry-Preserving Orthonormal Initialization for Low-Rank Adaptation in RLVR
Orthonormal initialization for LoRA in RLVR achieves the minimal gap to full fine-tuning, stabilizes training, and outperforms standard LoRA and prior variants on mathematical reasoning benchmarks.
-
Mechanism-Guided Selective Unlearning for RLVR-Induced Reasoning
MAST ranks attention-projection tensors by off-principal energy, update magnitude, and forget-gradient coupling to selectively unlearn RLVR-induced reasoning, achieving significant forgetting on MATH while preserving GSM8K and retain MATH unlike full-parameter updates.
-
Sparsity Curse: Understanding RLVR Model Parameter Space from Model Merging
RLVR induces sparse off-principal updates forming near-orthogonal shortcuts that degrade merging, addressed via Sensitivity-aware Resolving Merging using Fisher sensitivity, sparsification, and rescaling.
-
On the Scaling of PEFT: Towards Million Personal Models of Trillion Parameters
PEFT adapters are positioned as persistent personal state on foundation models, organized via Scale Up, Scale Down, and Scale Out axes, with MinT as an infrastructure example for managing them.
-
Pion: A Spectrum-Preserving Optimizer via Orthogonal Equivalence Transformation
Pion is an optimizer that preserves the singular values of weight matrices in LLM training by applying orthogonal equivalence transformations.
-
Learning to Foresee: Unveiling the Unlocking Efficiency of On-Policy Distillation
On-policy distillation gains efficiency from early foresight in module allocation and update directions, which the proposed EffOPD method exploits for 3x faster training with comparable performance.
-
SparseRL-Sync: Lossless Weight Synchronization with ~100x Less Communication
SparseRL-Sync achieves lossless weight synchronization in large-scale RL by sending only changed parameters, reducing communication volume by roughly 100x under observed 99%+ element-level sparsity.