The A.I. Daily
DeepSeek open-sources its DeepInfer inference engine
DeepSeek has open-sourced DeepInfer, the high-performance inference framework behind its large-scale production services, with this release focused on offline inference. It supports DeepSeek-V4.1-Flash, offers parallelism such as attention DP and MoE EP with EPLB, optimized kernels including FlashMLA and DeepGEMM, and DSpark draft-token decoding with confidence-scheduled verification.