Learning to Reason with Compressed Context: Ground-Truth-Free Adaptation of OmniLLMs via Self-Distillation

arXiv:2609.39953 · cs.CV · Submitted 2026-09-30 · Read on arXiv

cs.CV

Submitted: 2026-09-30

Updated: 2026-09-30

Code: https://github.com/Bamboos2003/CAFD

Terminology

• Accuracy — 61× in this paper · explained in 1 episode
• Benchmark — 58× in this paper · explained in 2 episodes
• Adaptation — 56× in this paper · explained in 1 episode
• Distillation — 35× in this paper · explained in 2 episodes
• Pruning — 14× in this paper · explained in 4 episodes
• Self-distillation — 13× in this paper · explained in 6 episodes
• Alignment — 5× in this paper · explained in 1 episode
• Conference on Computer Vision and Pattern Recognition — 4× in this paper
• On-Policy Distillation — 4× in this paper · explained in 3 episodes
• Affective Computing — 2× in this paper · explained in 1 episode
• Aggregation — 2× in this paper · explained in 1 episode
• Domain Adaptation — 2× in this paper · explained in 5 episodes
• Allocator — 1× in this paper · explained in 1 episode
• Calibration — 1× in this paper · explained in 7 episodes
• Causal Reasoning — 1× in this paper · explained in 2 episodes
• Consistency — 1× in this paper · explained in 3 episodes
• Consistency Distillation — 1× in this paper · explained in 2 episodes
• Exponential Moving Average (EMA) — 1× in this paper · explained in 2 episodes
• Fine-Tuning — 1× in this paper · explained in 5 episodes
• GRPO — 1× in this paper · explained in 6 episodes
• Generative AI — 1× in this paper · explained in 2 episodes
• Hallucination — 1× in this paper · explained in 3 episodes
• Instruction Tuning — 1× in this paper · explained in 2 episodes
• Redundancy — 1× in this paper · explained in 2 episodes
• Regularization — 1× in this paper · explained in 3 episodes
• Robustness — 1× in this paper · explained in 5 episodes
• Sparsity — 1× in this paper · explained in 4 episodes

Sources

• OmniVideo-100K: A Dataset for Audio-Visual Reasoning through Structured Scripts and Evidence Chains
• Thinking Without Images: Internalizing Visual Manipulation with On-Policy Self-Distillation
• OmniFocus: Query-Guided Modality-Balanced Token Compression for Omni-Modal Large Language Models
• AVOC: Enhancing Hour-Level Audio-Video Understanding in Omni-Modal LLMs via Retrieval-Inspired Token Compression
• OmniOPSD: Rationale-Privileged On-Policy Self-Distillation for Affective Computing
• OmniRefine: Alignment-Aware Cooperative Compression for Efficient Omnimodal Large Language Models
• VITA: Towards Open-Source Interactive Omni Multimodal LLM
• Allocation Before Ranking: Decoupled Token Compression for OmniLLMs
• OmniDelta: Skill-Driven Budget Allocation for Token Compression in OmniLLMs
• DASH: Dynamic Audio-Driven Semantic Chunking for Efficient Omnimodal Token Compression
• OmniDrop: Layer-wise Token Pruning for Omni-modal LLMs via Query-Guidance
• OmniScope: Modality-Decoupled Token Compression for Omnimodal Large Language Models
• OmniPack: Unified Token Compression for Efficient Omni-modal Large Language Models
• O-MARC: Omni Memory-Augmented Compression Distillation for Efficient Video Understanding
• Qwen2.5-Omni Technical Report
• Qwen3-Omni Technical Report
• OmniSelect: Dynamic Modality-Aware Token Compression for Efficient Omni-modal Large Language Models

Related papers

• Loss Knows Best: Detecting Annotation Errors in Videos via Loss Trajectories
• AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories
• Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift
• MambaX-Net: Dual-Input Mamba-Enhanced Cross-Attention Network for Longitudinal MRI Segmentation
• TeleOCR: Navigating Document Parsing Across Digital and Camera-Captured Documents
• A Survey on Efficient Vision-Language-Action Models