[1/2] Add FP8 Blockscale MoE CUTLASS kernel for Blackwell (#5281)
This commit is contained in:
7
sgl-kernel/python/sgl_kernel/__init__.py
Normal file → Executable file
7
sgl-kernel/python/sgl_kernel/__init__.py
Normal file → Executable file
@@ -41,7 +41,12 @@ from sgl_kernel.gemm import (
|
||||
sgl_per_token_group_quant_int8,
|
||||
sgl_per_token_quant_fp8,
|
||||
)
|
||||
from sgl_kernel.moe import moe_align_block_size, moe_fused_gate, topk_softmax
|
||||
from sgl_kernel.moe import (
|
||||
fp8_blockwise_scaled_grouped_mm,
|
||||
moe_align_block_size,
|
||||
moe_fused_gate,
|
||||
topk_softmax,
|
||||
)
|
||||
from sgl_kernel.sampling import (
|
||||
min_p_sampling_from_probs,
|
||||
top_k_renorm_prob,
|
||||
|
||||
Reference in New Issue
Block a user