""" LightLLM-optimized Text Encoder implementation Extracts core inference optimizations from LightLLM for LightX2V integration Available Encoders: 1. LightLLMServiceTextEncoder - 通过 HTTP 服务调用 LightLLM(需要独立服务) 2. LightLLMKernelTextEncoder - 基于 HuggingFace 模型 + Triton Kernels 优化 """ from .qwen25_text_encoder_kernel import LightLLMKernelTextEncoder from .qwen25_text_encoder_service import LightLLMServiceTextEncoder __all__ = [ "LightLLMServiceTextEncoder", # Service模式:通过HTTP调用LightLLM服务 "LightLLMKernelTextEncoder", # Kernel模式:HF模型 + Triton kernels ]