training-free framework
This framework refers to approaches where models can perform tasks without the traditional training phase, often relying on rules, heuristics, or adaptable mechanisms to operate effectively immediately upon deployment.
- Each Complexity Deserves a Pruning Policy
- FlexAC: Towards Flexible Control of Associative Reasoning in Multimodal Large Language Models
- Latent Refinement via Flow Matching for Training-free Linear Inverse Problem Solving
- Open-Vocabulary Part Segmentation via Progressive and Boundary-Aware Strategy
- Personalized Safety in LLMs: A Benchmark and A Planning-Based Agent Approach
- SceneDecorator: Towards Scene-Oriented Story Generation with Scene Planning and Scene Consistency
- SolverLLM: Leveraging Test-Time Scaling for Optimization Problem via LLM-Guided Search
- Sparse VideoGen2: Accelerate Video Generation with Sparse Attention via Semantic-Aware Permutation
- SpecEM: Training-Free LLM Ensembling via Iterative Drafting, Verification, and Online Feedback
- Unleashing Diffusion Transformers for Visual Correspondence by Modulating Massive Activations
- When Semantics Mislead Vision: Mitigating Large Multimodal Models Hallucinations in Scene Text Spotting and Understanding
- Wukong's 72 Transformations: High-fidelity Textured 3D Morphing via Flow Models