Attention Sinks Are Provably Necessary in Softmax Transformers: Evidence from Trigger-Conditional Tasks
Paper
• 2603.11487 • Published
• 1
Kye Gomez
kye
AI & ML interests
Neuroscience, Behavior Science, Anti-Matter, Anti-Gravity propulsion,
Recent Activity
upvoted a paper about 4 hours ago
Attention Sinks Are Provably Necessary in Softmax Transformers: Evidence from Trigger-Conditional Tasks upvoted a paper about 4 hours ago
Meta-Reinforcement Learning with Self-Reflection for Agentic Search upvoted a paper about 4 hours ago
Examining Reasoning LLMs-as-Judges in Non-Verifiable LLM Post-Training Organizations
upvoted 10 papers about 4 hours ago
Meta-Reinforcement Learning with Self-Reflection for Agentic Search
Paper
• 2603.11327 • Published
• 3
Examining Reasoning LLMs-as-Judges in Non-Verifiable LLM Post-Training
Paper
• 2603.12246 • Published
• 4
Are Video Reasoning Models Ready to Go Outside?
Paper
• 2603.10652 • Published
• 6
EndoCoT: Scaling Endogenous Chain-of-Thought Reasoning in Diffusion Models
Paper
• 2603.12252 • Published
• 9
One Model, Many Budgets: Elastic Latent Interfaces for Diffusion Transformers
Paper
• 2603.12245 • Published
• 11
DreamVideo-Omni: Omni-Motion Controlled Multi-Subject Video Customization with Latent Identity Reinforcement Learning
Paper
• 2603.12257 • Published
• 24
IndexCache: Accelerating Sparse Attention via Cross-Layer Index Reuse
Paper
• 2603.12201 • Published
• 32
Video-Based Reward Modeling for Computer-Use Agents
Paper
• 2603.10178 • Published
• 30
Spatial-TTT: Streaming Visual-based Spatial Intelligence with Test-Time Training
Paper
• 2603.12255 • Published
• 63
upvoted 5 papers 1 day ago
ReMix: Reinforcement routing for mixtures of LoRAs in LLM finetuning
Paper
• 2603.10160 • Published
• 20
Causal Concept Graphs in LLM Latent Space for Stepwise Reasoning
Paper
• 2603.10377 • Published
• 3
UniCom: Unified Multimodal Modeling via Compressed Continuous Semantic Representations
Paper
• 2603.10702 • Published
• 3
LLM2Vec-Gen: Generative Embeddings from Large Language Models
Paper
• 2603.10913 • Published
• 29
Flash-KMeans: Fast and Memory-Efficient Exact K-Means
Paper
• 2603.09229 • Published
• 62
upvoted 2 papers 2 days ago
upvoted 3 papers 3 days ago
MM-Zero: Self-Evolving Multi-Model Vision Language Models From Zero Data
Paper
• 2603.09206 • Published
• 41
Fish Audio S2 Technical Report
Paper
• 2603.08823 • Published
• 22
Omni-Diffusion: Unified Multimodal Understanding and Generation with Masked Discrete Diffusion
Paper
• 2603.06577 • Published
• 43