Lightning OPD: Efficient Post-Training for Large Reasoning Models with Offline On-Policy Distillation
Paper • 2604.13010 • Published • 10
hassenhamdi
hassenhamdi
AI & ML interests
None yet
Recent Activity
commentedon a paper about 7 hours ago
GlobalSplat: Efficient Feed-Forward 3D Gaussian Splatting via Global Scene Tokens commentedon a paper about 23 hours ago
Three-Phase Transformer commentedon a paper about 23 hours ago
Three-Phase TransformerOrganizations
upvoted 3
papers 3 days ago
upvoted 5
papers 8 days ago
The Master Key Hypothesis: Unlocking Cross-Model Capability Transfer via Linear Subspace Alignment
Paper • 2604.06377 • Published • 7
Flux Attention: Context-Aware Hybrid Attention for Efficient LLMs Inference
Paper • 2604.07394 • Published • 16
DMax: Aggressive Parallel Decoding for dLLMs
Paper • 2604.08302 • Published • 50
Rethinking Generalization in Reasoning SFT: A Conditional Analysis on Optimization, Data, and Model Capability
Paper • 2604.06628 • Published • 316
Generation Models Know Space: Unleashing Implicit 3D Priors for Scene Understanding
Paper • 2603.19235 • Published • 95
upvoted 9
papers 9 days ago
Efficient Reasoning on the Edge
Paper • 2603.16867 • Published • 21
Make LoRA Great Again: Boosting LoRA with Adaptive Singular Values and Mixture-of-Experts Optimization Alignment
Paper • 2502.16894 • Published • 33
Learning to Reason without External Rewards
Paper • 2505.19590 • Published • 31
Fast Spatial Memory with Elastic Test-Time Training
Paper • 2604.07350 • Published • 7
The Depth Ceiling: On the Limits of Large Language Models in Discovering Latent Planning
Paper • 2604.06427 • Published • 11
Neural Computers
Paper • 2604.06425 • Published • 29
TC-AE: Unlocking Token Capacity for Deep Compression Autoencoders
Paper • 2604.07340 • Published • 16
SEVerA: Verified Synthesis of Self-Evolving Agents
Paper • 2603.25111 • Published • 31
MARS: Enabling Autoregressive Models Multi-Token Generation
Paper • 2604.07023 • Published • 38
upvoted 3
papers 12 days ago