Mojo struct
MLASparseConfig
struct MLASparseConfig[qkv_dtype: DType]
Fieldsβ
- βnum_q_heads (
Int): - βnum_kv_heads (
Int): - βqk_depth (
Int): - βv_depth (
Int): - βtop_k (
Int): - βgroup (
Int):
Implemented traitsβ
AnyType,
ImplicitlyDestructible
comptime membersβ
B_TOPKβ
comptime B_TOPK = 128
cta_groupβ
comptime cta_group = 2
k_swizzle_modeβ
comptime k_swizzle_mode = TensorMapSwizzle.SWIZZLE_128B
num_threadsβ
comptime num_threads = 512
output_swizzle_modeβ
comptime output_swizzle_mode = TensorMapSwizzle.SWIZZLE_128B
q_smem_depthβ
comptime q_smem_depth = 192
q_swizzle_modeβ
comptime q_swizzle_mode = TensorMapSwizzle.SWIZZLE_128B
q_tmem_depthβ
comptime q_tmem_depth = 384
qkv_dtype_sizeβ
comptime qkv_dtype_size = size_of[qkv_dtype]()
sm100_tmem_colsβ
comptime sm100_tmem_cols = 512
Methodsβ
__init__β
__init__(out self, *, num_q_heads: Int, num_kv_heads: Int, qk_depth: Int, v_depth: Int, top_k: Int, group: Int)
Was this page helpful?
Thank you! We'll create more content like this.
Thank you for helping us improve!