vllm.v1.attention.ops.turboquant_soa ¶
Modules:
-
triton_turboquant_decode–Triton fused TurboQuant decode attention.
-
triton_turboquant_decode_v2–Optimized Triton TurboQuant decode attention (v2).
-
triton_turboquant_store–Fused Triton kernels for TurboQuant KV store.
-
triton_turboquant_unified_attention–Unified (prefill + decode) Triton attention kernel for TurboQuant.