π Day 2 of #OpenSourceWeek: DeepEP
Excited to introduce DeepEP - the first open-source EP communication library for MoE model training and inference.
β Efficient and optimized all-to-all communication
β Both intranode and internode support with NVLink and RDMA
β High-throughput kernels for training and inference prefilling
β Low-latency kernels for inference decoding
β Native FP8 dispatch support
β Flexible GPU resource control for computation-communication overlapping
π GitHub: https://github.com/deepseek-ai/DeepEP
@ImmersiveAi
Excited to introduce DeepEP - the first open-source EP communication library for MoE model training and inference.
β Efficient and optimized all-to-all communication
β Both intranode and internode support with NVLink and RDMA
β High-throughput kernels for training and inference prefilling
β Low-latency kernels for inference decoding
β Native FP8 dispatch support
β Flexible GPU resource control for computation-communication overlapping
π GitHub: https://github.com/deepseek-ai/DeepEP
@ImmersiveAi