https://uncensoredainews.com/news/new-4-bit-adamw-quantization-cuts-validation-loss-gap-in-local-llm-pretraining-7262026-10-09T15:16:46ZUncensored AI Newsen2026-10-09T15:16:46ZPreprint Redesigns 4-bit AdamW Quantization via Preconditioner-Space Roundinghttps://uncensoredainews.com/news/sapd-new-off-policy-distillation-aligns-privileged-guidance-to-reasoning-steps-6842026-10-09T12:15:57ZUncensored AI Newsen2026-10-09T12:15:57ZSAPD Preprint: Step-Aligned Privileged Distillation Enables Rollout-Free LLM Post-Traininghttps://uncensoredainews.com/news/disparq-learns-discrete-part-concepts-for-interpretable-vision-models-without-labels-or-language-6832026-10-09T09:14:42ZUncensored AI Newsen2026-10-09T09:14:42ZDisParQ Preprint Learns Discrete Part Concepts from Vision Backbones Without Labels or Languagehttps://uncensoredainews.com/news/tr-ptq-delivers-integer-only-transformer-quantization-with-under-1-5-accuracy-drop-6802026-10-09T00:11:49ZUncensored AI Newsen2026-10-09T00:11:49ZTR-PTQ Reformulates Transformer Quantization for Integer-Only Inferencehttps://uncensoredainews.com/news/uniskill-aligns-skill-proposals-to-evolving-llm-agents-without-extra-rollouts-6792026-10-08T21:10:52ZUncensored AI Newsen2026-10-08T21:10:52ZUniSkill Preprint Proposes Shared-Policy Skill Learning for LLM Agentshttps://uncensoredainews.com/news/llm-agents-build-transparent-open-source-multiphysics-models-for-electrochemical-devices-6782026-10-08T18:09:46ZUncensored AI Newsen2026-10-08T18:09:46ZLLM Agents Generate Open-Source Multiphysics Models from Human Specificationshttps://uncensoredainews.com/news/sldr-defense-uses-signed-safety-sensitivity-to-counter-malicious-fine-tuning-6772026-10-08T15:09:10ZUncensored AI Newsen2026-10-08T15:09:10ZSLDR Preprint Uses Signed Safety Sensitivity for Targeted Post-Fine-Tuning Defensehttps://uncensoredainews.com/news/residualquant-preprint-targets-kv-cache-in-looped-transformers-with-2-bit-residuals-6762026-10-08T12:07:51ZUncensored AI Newsen2026-10-08T12:07:51ZResidualQuant Preprint Proposes 2-Bit KV Cache Quantization for Looped Transformershttps://uncensoredainews.com/news/sigma-preprint-shows-llms-can-self-improve-alignment-from-a-model-spec-6232026-10-08T09:06:37ZUncensored AI Newsen2026-10-08T09:06:37ZSIGMA Preprint Shows LLMs Can Self-Improve Alignment from a Model Spechttps://uncensoredainews.com/news/preprint-diagnoses-attenuated-dynamics-in-time-series-foundation-models-6222026-10-08T06:05:40ZUncensored AI Newsen2026-10-08T06:05:40ZPreprint Diagnoses Attenuated Dynamics in Time-Series Foundation Modelshttps://uncensoredainews.com/news/training-free-two-stage-low-rank-compensation-improves-extremely-quantized-llms-6212026-10-08T03:05:10ZUncensored AI Newsen2026-10-08T03:05:10ZAlign-Then-Correct: Training-Free Two-Stage Compensation for 2-Bit LLMshttps://uncensoredainews.com/news/osfp4-preprint-improves-accuracy-for-nvfp4-quantization-via-joint-smoothing-and-scaling-6202026-10-08T00:03:59ZUncensored AI Newsen2026-10-08T00:03:59ZOSFP4 Preprint Proposes Joint Smoothing and Scaling Optimization for NVFP4 Quantizationhttps://uncensoredainews.com/news/new-framework-helps-ai-simulate-realistic-therapy-clients-without-drifting-6192026-10-07T21:03:02ZUncensored AI Newsen2026-10-07T21:03:02ZMASC Framework Aims to Reduce Persona Drift in AI Client Simulations for Counselinghttps://uncensoredainews.com/news/ai2-publishes-byte-level-model-retrofits-in-nature-releases-new-checkpoints-6582026-10-07T18:01:34ZUncensored AI Newsen2026-10-07T18:01:34ZAi2 Publishes Byte-Level Model Retrofits in Nature, Releases New Checkpoints