Search papers, labs, and topics across Lattice.
2
0
5
2
Stop forcing your multimodal encoders to inherit suboptimal LLM parallelism strategies: heterogeneous parallelism unlocks up to 49% higher TFLOPS/GPU.
Multimodal models can now achieve state-of-the-art performance in real-world tasks like document understanding and audio-video comprehension with significantly reduced inference latency thanks to novel token-reduction techniques.
