标签
SwiftLM is a Swift-native LLM inference server for Apple Silicon that runs large models without Python, using SSD streaming to load MoE weights and enabling 122B models on 64 GB Macs.