flashinfer.fused_moe.PrimsTsConfig

class flashinfer.fused_moe.PrimsTsConfig

Explicit Prims-TS backend for SM100 and SM103.

Opt-in only: this config is never part of the default backend list. The physical weight and activation layouts are exactly those produced by TrtllmFp4Config (NVFP4×NVFP4, MajorK) and TrtllmBf16Config (BF16×BF16, BlockMajorK). Register the same dictionary under "prims_ts" and the matching TRT-LLM key to share one quantized representation between the two runners.

The GEMM middle stage is Prims-TS; routing and finalize stay on the TRT-LLM Gen path.

__init__() → None

Methods

__init__()
prepare_activations(hidden_states_bf16, *, quant)	Build activations for the matching TRTLLM physical layout.
prepare_weights(w1_bf16, w2_bf16, *, quant, ...)	Build the shared TRTLLM physical weight view for Prims-TS.
supported(arch)