[chore] Remove unused ep_moe cuda kernels (#9956)

sgl-kernel/python/sgl_kernel/__init__.py

@@ -77,9 +77,6 @@ from sgl_kernel.memory import set_kv_buffer_kernel
 from sgl_kernel.moe import (
     apply_shuffle_mul_sum,
     cutlass_fp4_group_mm,
-    ep_moe_post_reorder,
-    ep_moe_pre_reorder,
-    ep_moe_silu_and_mul,
     fp8_blockwise_scaled_grouped_mm,
     moe_align_block_size,
     moe_fused_gate,

sgl-kernel/python/sgl_kernel/moe.py

@@ -71,70 +71,6 @@ def moe_fused_gate(
     )
 
 
-def ep_moe_pre_reorder(
-    input_tensor,
-    gateup_input,
-    src2dst,
-    topk_ids,
-    a1_scales,
-    start_expert_id,
-    end_expert_id,
-    topk,
-    use_per_token_if_dynamic,
-):
-    return torch.ops.sgl_kernel.ep_moe_pre_reorder.default(
-        input_tensor,
-        gateup_input,
-        src2dst,
-        topk_ids,
-        a1_scales,
-        start_expert_id,
-        end_expert_id,
-        topk,
-        use_per_token_if_dynamic,
-    )
-
-
-def ep_moe_silu_and_mul(
-    gateup_output,
-    down_input,
-    reorder_topk_ids,
-    scales,
-    start_expert_id,
-    end_expert_id,
-):
-    return torch.ops.sgl_kernel.ep_moe_silu_and_mul.default(
-        gateup_output,
-        down_input,
-        reorder_topk_ids,
-        scales,
-        start_expert_id,
-        end_expert_id,
-    )
-
-
-def ep_moe_post_reorder(
-    down_output,
-    output,
-    src2dst,
-    topk_ids,
-    topk_weights,
-    start_expert_id,
-    end_expert_id,
-    topk,
-):
-    return torch.ops.sgl_kernel.ep_moe_post_reorder.default(
-        down_output,
-        output,
-        src2dst,
-        topk_ids,
-        topk_weights,
-        start_expert_id,
-        end_expert_id,
-        topk,
-    )
-
-
 def fp8_blockwise_scaled_grouped_mm(
     output,
     a_ptrs,