Add moe topk softmax templated from vllm (#4302)
This commit is contained in:
@@ -33,7 +33,7 @@ from sgl_kernel.gemm import (
|
||||
sgl_per_token_group_quant_fp8,
|
||||
sgl_per_token_quant_fp8,
|
||||
)
|
||||
from sgl_kernel.moe import moe_align_block_size
|
||||
from sgl_kernel.moe import moe_align_block_size, topk_softmax
|
||||
from sgl_kernel.sampling import (
|
||||
min_p_sampling_from_probs,
|
||||
top_k_renorm_prob,
|
||||
|
||||
Reference in New Issue
Block a user