Search papers, labs, and topics across Lattice.
Open-Sora Plan Team, Peking University
3
0
8
Achieve near state-of-the-art video generation quality with OSP-Next, while realizing up to 2.27x speedups on Ascend 950PR GPUs through a novel combination of sparse attention, quantization, and parallelism techniques.
MiniMax-M2 proves that massive parameter counts don't always translate to better agentic performance; strategic activation of a smaller subset can unlock frontier-level intelligence.
Rigid reward clipping throws away valuable information just beyond the boundary, but a simple stochastic rescue of these signals can substantially boost RLVR performance.
