Search papers, labs, and topics across Lattice.
Affiliation:
4
0
8
MDD achieves up to 2.95x faster video generation while preserving visual fidelity by decoupling magnitude and directional components in the denoising process.
Real-time semantic corrections during video generation can drastically enhance visual fidelity and coherence without retraining the model.
SMEPilot achieves up to 3.94脳 faster LLM inference by intelligently balancing workloads between CPU and SME units based on operator characteristics.
Gemini Embedding 2's unified multimodal embeddings beat specialized models across diverse tasks and even generalize zero-shot to niche fields like astronomy and culinary arts.
