flashinfer.fused_moe.ExecutionConfig¶

class flashinfer.fused_moe.ExecutionConfig(enable_pdl: bool | None = None, tune_max_num_tokens: int = 8192)¶

Runtime execution parameters.

Parameters:

enable_pdl (bool or None) – Persistent device launch. None → auto (True for sm90+).
tune_max_num_tokens (int) – Token budget hint for autotuner / CUDA graph capture.

__init__(enable_pdl: bool | None = None, tune_max_num_tokens: int = 8192) → None¶

Methods

__init__([enable_pdl, tune_max_num_tokens])

Attributes

`enable_pdl`
`tune_max_num_tokens`