vllm.models.muse_spark.amd.ops
¶
Modules:
-
fused_markov–The Muse Spark DSpark Markov rounds with one launch a round (--monokernel).
-
fused_norm_expand–Pre-expert RMSNorm written once per route, in the
[T * top_k, H]layout -
router_bf16x3–FP32-exact router GEMM on bf16 matrix cores.