vllm.models.muse_spark.nvidia.specialized_decode.cutedsl.draft
¶
Modules:
-
context–The Muse Spark DSpark draft's context K/V chain around its K/V projection,
-
input_norm–The DSpark draft's first norm (its first layer's pre_qkv_proj_norm of the
-
markov–CuTe DSL kernels of the fused DSpark Markov sampling loop.