serving

Modules

modelopt.torch.kernels.quantization.linear_attention.serving.forward	Native vLLM 0.30 Triton forwards; training owns private, differentiable state.
modelopt.torch.kernels.quantization.linear_attention.serving.replay	Training adapter for the optional INT8/Hadamard quantized-ReplaySSM fork.

Training adapter for native vLLM linear-attention forward kernels.

ModelOpt owns the training state, QDQ and backward. This package imports the selected serving arithmetic; it does not modify the vLLM engine or scheduler.