spikingjelly.activation_based.distributed package#

The root package contains SNN-specific tensor-parallel primitives. Training and offline inference interfaces are grouped by workload: native PyTorch vision lives in distributed.vision and Megatron Core language-model execution lives in distributed.llm. The package also provides a model-independent managed SGLang offline Engine and topology-independent artifact exporter; concrete model recipes own their weight mappings and external runtime packages. Importing the root package imports neither optional runtime.

Tensor parallel primitives#

Vision training and inference#

LLM training and inference#