verl/workers/sharding_manager/fsdp_vllm.py

• FSDPVLLMShardingManager · class · L34-L133 — class FSDPVLLMShardingManager(BaseShardingManager)
• init · method · L36-L67 — def init(self, module: FSDP, inference_engine: LLM, model_config, full_params: bool = False, device_mesh: DeviceMesh = None)
• enter · method · L69-L91 — def enter(self)
• exit · method · L93-L110 — def exit(self, exc_type, exc_value, traceback)
• preprocess_data · method · L112-L119 — def preprocess_data(self, data: DataProto) -> DataProto: # TODO: Current impl doesn't consider FSDP with torch micro-dp
• postprocess_data · method · L121-L133 — def postprocess_data(self, data: DataProto) -> DataProto: # TODO: Current impl doesn't consider FSDP with torch micro-dp