flashinfer.topk.can_implement_filtered_topk

flashinfer.topk.can_implement_filtered_topk() → bool

Check if the GPU supports enough shared memory for FilteredTopK algorithm.

FilteredTopK requires 128KB dynamic shared memory. This function checks if the current GPU’s max shared memory per SM is sufficient.

Returns:

True if GPU supports FilteredTopK, False otherwise.

Return type:

bool