1. Update MOE; 2. Update sglang mHC; 3. Update test scripts; 4 Add new

ops.

1. Update MOE; 2. Update sglang mHC; 3. Update test scripts; 4 Add new
ops.
bb596f6e · xiaowei.zhang · d9ebb683 · bb596f6e · bb596f6e · bb596f6e
Commit bb596f6e authored Jun 04, 2026 by xiaowei.zhang
20 changed files
--- a/aiter/ops/tilelang/mhc/pre_norm_fn_splitk_kernel.py
+++ b/aiter/ops/tilelang/mhc/pre_norm_fn_splitk_kernel.py
--- a/aiter/ops/triton/configs/BW200B-EXTEND_ATTENTION-V2-DECODE-FP16.json
+++ b/aiter/ops/triton/configs/BW200B-EXTEND_ATTENTION-V2-DECODE-FP16.json
--- a/aiter/ops/triton/configs/BW200B-EXTEND_ATTENTION-V2-FP16.json
+++ b/aiter/ops/triton/configs/BW200B-EXTEND_ATTENTION-V2-FP16.json
--- a/aiter/ops/triton/configs/chunk_fwd_o/chunk_fwd_o-gfx936.json
+++ b/aiter/ops/triton/configs/chunk_fwd_o/chunk_fwd_o-gfx936.json
--- a/aiter/ops/triton/configs/chunk_fwd_o/chunk_fwd_o-gfx938.json
+++ b/aiter/ops/triton/configs/chunk_fwd_o/chunk_fwd_o-gfx938.json
--- a/aiter/ops/triton/configs/chunk_gated_delta_rule_fwd_h/chunk_gated_delta_rule_fwd_h-gfx936.json
+++ b/aiter/ops/triton/configs/chunk_gated_delta_rule_fwd_h/chunk_gated_delta_rule_fwd_h-gfx936.json
--- a/aiter/ops/triton/configs/chunk_gated_delta_rule_fwd_h/chunk_gated_delta_rule_fwd_h-gfx938.json
+++ b/aiter/ops/triton/configs/chunk_gated_delta_rule_fwd_h/chunk_gated_delta_rule_fwd_h-gfx938.json
--- a/aiter/ops/triton/configs/extend_attn/_fwd_kernel_v2-device=gfx938_cu72.json
+++ b/aiter/ops/triton/configs/extend_attn/_fwd_kernel_v2-device=gfx938_cu72.json
--- a/aiter/ops/triton/configs/extend_attn/_fwd_kernel_v2_decode-device=gfx938_cu72.json
+++ b/aiter/ops/triton/configs/extend_attn/_fwd_kernel_v2_decode-device=gfx938_cu72.json
--- a/aiter/ops/triton/configs/fused_recurrent_gated_delta_rule_packed_decode/fused_recurrent_gated_delta_rule_packed_decode-gfx936.json
+++ b/aiter/ops/triton/configs/fused_recurrent_gated_delta_rule_packed_decode/fused_recurrent_gated_delta_rule_packed_decode-gfx936.json
--- a/aiter/ops/triton/configs/fused_recurrent_gated_delta_rule_packed_decode/fused_recurrent_gated_delta_rule_packed_decode-gfx938.json
+++ b/aiter/ops/triton/configs/fused_recurrent_gated_delta_rule_packed_decode/fused_recurrent_gated_delta_rule_packed_decode-gfx938.json
--- a/aiter/ops/triton/configs/fused_sigmoid_gating_delta_rule_update/fused_sigmoid_gating_delta_rule_update-gfx936.json
+++ b/aiter/ops/triton/configs/fused_sigmoid_gating_delta_rule_update/fused_sigmoid_gating_delta_rule_update-gfx936.json
--- a/aiter/ops/triton/configs/fused_sigmoid_gating_delta_rule_update/fused_sigmoid_gating_delta_rule_update-gfx938.json
+++ b/aiter/ops/triton/configs/fused_sigmoid_gating_delta_rule_update/fused_sigmoid_gating_delta_rule_update-gfx938.json
--- a/aiter/ops/triton/configs/fused_sigmoid_gating_delta_rule_update_recurrent/fused_sigmoid_gating_delta_rule_update_recurrent-gfx936.json
+++ b/aiter/ops/triton/configs/fused_sigmoid_gating_delta_rule_update_recurrent/fused_sigmoid_gating_delta_rule_update_recurrent-gfx936.json
--- a/aiter/ops/triton/configs/fused_sigmoid_gating_delta_rule_update_recurrent/fused_sigmoid_gating_delta_rule_update_recurrent-gfx938.json
+++ b/aiter/ops/triton/configs/fused_sigmoid_gating_delta_rule_update_recurrent/fused_sigmoid_gating_delta_rule_update_recurrent-gfx938.json
--- a/aiter/ops/triton/configs/moe/E=160,N=320,device_name=BW200B,dtype=fp8_w8a8,is_bottom=True.json
+++ b/aiter/ops/triton/configs/moe/E=160,N=320,device_name=BW200B,dtype=fp8_w8a8,is_bottom=True.json
--- a/aiter/ops/triton/configs/moe/E=160,N=320,device_name=BW200B,dtype=fp8_w8a8.json
+++ b/aiter/ops/triton/configs/moe/E=160,N=320,device_name=BW200B,dtype=fp8_w8a8.json
--- a/aiter/ops/triton/configs/moe/E=160,N=320,device_name=BW200B,dtype=int8_w8a8,is_bottom=True.json
+++ b/aiter/ops/triton/configs/moe/E=160,N=320,device_name=BW200B,dtype=int8_w8a8,is_bottom=True.json
--- a/aiter/ops/triton/configs/moe/E=160,N=320,device_name=BW200B,dtype=int8_w8a8.json
+++ b/aiter/ops/triton/configs/moe/E=160,N=320,device_name=BW200B,dtype=int8_w8a8.json
--- a/aiter/ops/triton/configs/moe/E=256,N=128,device_name=K100_AI,dtype=int8_w8a8,block_shape=[128,128].json
+++ b/aiter/ops/triton/configs/moe/E=256,N=128,device_name=K100_AI,dtype=int8_w8a8,block_shape=[128,128].json