vllm.kernels.triton

Triton kernel implementations.

Modules:

• qkv_padded_fp8_quant –

  Stride-aware FP8 quantization with head_dim padding for ViT attention.