Second-order screen predicts quantization mismatch on five language models
On unseen Mistral-7B-v0.3, all six predictions fell in band and vLLM throughput was predicted within 0.4%.
Stories per day · 30 days
Relateddistillation1distributionally-robust-optimization1fisher-information1inference1language-models1mathematical-reasoning1model-training1observation-coordination1quantization1scenario-generation1stochastic-programming1
10 stories, newest first · one card per story
Stories tagged #kl-divergence
OC-OPD improves on-policy distillation via observation-coordinated supervision
OC-OPD uses reverse and forward KL divergences with a four-observation matrix to coordinate gradients, achieving strong gains in mathematical reasoning and OOD transfer.
KL-DRO centered on a learned density uses adaptive scenarios
On scarce records, a learned nominal lowered cost under distribution shift on newsvendor and grid capacity problems.
KL Before and After Mixing: A Round-Complexity Separation in Parallel Sampling
Theory paper separates two KL measures for parallel sampling, with very different round costs.
SimpleMark: Fast Multi-Bit Text Watermarking under f -Divergence Constraints
SimpleMark embeds multi-bit text watermarks while bounding f-divergence from the base language model.
On-Policy or Off-Policy Learning? A Systematic Study of Distillation Dynamics
Controlled distillation finds KL direction matters more than on-policy versus off-policy rollouts.
Do We Really Need KL Divergence for On-Policy Distillation of Large Language Models?
Researchers show on-policy LLM distillation works without KL divergence if updates move toward the teacher.
Minimally Invasive Steering of Language Models
MISVO steers frozen language models toward a test-time reward while limiting distribution shift via local KL geometry.
Safe Error Correction for Language Models: Frozen-Base Adjustment with Capability Preservation
A 34M logit corrector on frozen Gemma 4 E2B fixes 53.3% of exam errors without capability loss.
A Unified and Constrained View of Regularization-Based Robust Reinforcement Learning
Paper unifies regularization-based robust RL methods via new performance-gap upper bounds and jointly learned Lagrange multipliers.