Commit cbeaf0b5 authored by zhuwenwen's avatar zhuwenwen
Browse files

update fused_moe.py

parent 2b47d4fa
...@@ -918,9 +918,7 @@ def invoke_fused_moe_kernel(A: torch.Tensor, ...@@ -918,9 +918,7 @@ def invoke_fused_moe_kernel(A: torch.Tensor,
assert B_scale is None assert B_scale is None
EM = sorted_token_ids.shape[0] EM = sorted_token_ids.shape[0]
if use_int4_w4a16: if A.shape[0] < config["BLOCK_SIZE_M"]:
EM = sorted_token_ids.shape[0]
elif A.shape[0] < config["BLOCK_SIZE_M"]:
# optimize for small batch_size. # optimize for small batch_size.
# We assume that top_ids of each token is unique, so # We assume that top_ids of each token is unique, so
# so num_valid_experts <= batch_size <= BLOCK_SIZE_M, # so num_valid_experts <= batch_size <= BLOCK_SIZE_M,
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment