[MoE Refactor] [NPU] Refactor Ascend MoE implementation to reduce code duplication and align with community design (#25663)

Co-authored-by: ronnie_zheng <zl19940307@163.com>

docs_new/docs/hardware-platforms/ascend-npus/ascend_npu_optimization.mdx

@@ -304,7 +304,7 @@ Speculative decoding reduces per-token latency by generating draft tokens that a
   <tbody>
     <tr>
       <td>`--quantization modelslim`</td>
-      <td>Uses the Ascend ModelSlim quantization tool to load W8A8 pre-quantized model weights. This reduces model weight footprint by approximately 50% compared to BF16, allowing larger models to fit in NPU memory with minimal accuracy degradation. The quantization method is auto-detected from the model's `quant_model_description.json` file.</td>
+      <td>Uses the Ascend ModelSlim quantization tool to load W4A4/W4A8/W8A8/W4A16 pre-quantized model weights. This reduces model weight footprint by approximately 50% (for w8) or 75% (for w4) compared to BF16, allowing larger models to fit in NPU memory with minimal accuracy degradation. The quantization method is auto-detected from the model's `quant_model_description.json` file.</td>
       <td>`modelslim`</td>
     </tr>
   </tbody>
@@ -357,11 +357,6 @@ The following environment variables are used in other best practice configuratio
       <td>Configures the HCCL communication algorithm scheduling to use AIV (Ascend Intelligent Vision) expansion mode, which can improve communication efficiency for certain collective operations.</td>
       <td>Used in Qwen MoE and R1 non-DSA configurations</td>
     </tr>
-    <tr>
-      <td>`SGLANG_NPU_FUSED_MOE_MODE`</td>
-      <td>Controls the fused MoE optimization mode on Ascend NPU. `1` is default; `2` enables a more aggressive fusion strategy (`DISPATCH_FFN_COMBINE`) that can improve MoE dispatch throughput. Mode `2` requires `--quantization modelslim`. Used primarily with DeepSeek R1 models.</td>
-      <td>`1` or `2`</td>
-    </tr>
     <tr>
       <td>`SGLANG_NPU_USE_MLAPO=1`</td>
       <td><strong>(DeepSeek MLA-specific)</strong> Adopts the `MLAPO` fusion operator in the MLA (Multi-Head Latent Attention) preprocessing stage for DeepSeek models with MLA architecture.</td>

docs_new/docs/hardware-platforms/ascend-npus/best_practice/deepseek_r1.mdx

@@ -611,7 +611,6 @@ do
         export HCCL_BUFFSIZE=800
         export HCCL_SOCKET_IFNAME=<network-interface>
         export SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK=131072
-        export SGLANG_NPU_FUSED_MOE_MODE=2
         export SGLANG_USE_AG_AFTER_QLORA=1
         export TASK_QUEUE_ENABLE=2
 
@@ -625,13 +624,12 @@ do
         --nnodes 1 \
         --tp-size 16 \
         --mem-fraction-static 0.778 \
-        --quantization modelslim \
         --max-running-requests 16 \
         --disable-radix-cache \
         --chunked-prefill-size -1 \
         --max-prefill-tokens 60000 \
         --moe-a2a-backend ascend_fuseep \
-        --deepep-mode normal \
+        --fuseep-mode 2 \
         --speculative-algorithm NEXTN \
         --speculative-num-steps 1 \
         --speculative-eagle-topk 1 \
@@ -664,7 +662,6 @@ do
         export SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK=64
         export SGLANG_ENABLE_OVERLAP_PLAN_STREAM=1
         export SGLANG_LM_HEAD_TP=8
-        export SGLANG_NPU_FUSED_MOE_MODE=1
         export TASK_QUEUE_ENABLE=1
 
         python3 -m sglang.launch_server \
@@ -679,8 +676,8 @@ do
         --dp-size 32 \
         --mem-fraction-static 0.82 \
         --max-running-requests 1024 \
-        --quantization modelslim \
         --moe-a2a-backend ascend_fuseep \
+        --fuseep-mode 1 \
         --enable-dp-attention \
         --deepep-mode low_latency \
         --moe-dense-tp 1 \

docs_new/docs/hardware-platforms/ascend-npus/best_practice/minimax_m2_5.mdx

@@ -192,7 +192,6 @@ export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
 export SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK=140000
 export SGLANG_ENABLE_OVERLAP_PLAN_STREAM=1
 export SGLANG_EXTERNAL_MODEL_PACKAGE=custom_eagle3
-export SGLANG_NPU_FUSED_MOE_MODE=2
 export SGLANG_SET_CPU_AFFINITY=1
 export STREAMS_PER_DEVICE=32
 export TASK_QUEUE_ENABLE=1
@@ -211,8 +210,7 @@ python3 -m sglang.launch_server \
     --max-prefill-tokens 67000 \
     --cuda-graph-bs 2 4 8 12 16 18 20 22 24 26 \
     --moe-a2a-backend ascend_fuseep \
-    --deepep-mode auto \
-    --quantization modelslim \
+    --fuseep-mode 2 \
     --speculative-algorithm EAGLE3 \
     --speculative-draft-model-path $DRAFT_MODEL_PATH \
     --speculative-num-steps 3 \
@@ -300,7 +298,6 @@ export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
 export SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK=160000
 export SGLANG_ENABLE_OVERLAP_PLAN_STREAM=1
 export SGLANG_EXTERNAL_MODEL_PACKAGE=custom_eagle3
-export SGLANG_NPU_FUSED_MOE_MODE=2
 export SGLANG_SET_CPU_AFFINITY=1
 export STREAMS_PER_DEVICE=32
 export TASK_QUEUE_ENABLE=1
@@ -321,8 +318,7 @@ python3 -m sglang.launch_server \
     --max-prefill-tokens 80000 \
     --cuda-graph-bs 2 4 6 8 \
     --moe-a2a-backend ascend_fuseep \
-    --deepep-mode auto \
-    --quantization modelslim \
+    --fuseep-mode 2 \
     --speculative-algorithm EAGLE3 \
     --speculative-draft-model-path $DRAFT_MODEL_PATH \
     --speculative-num-steps 3 \

docs_new/docs/hardware-platforms/ascend-npus/best_practice/qwen3_235b_a22b.mdx

@@ -174,7 +174,7 @@ export PYTORCH_NPU_ALLOC_CONF=expandable_segments:True
 export SGLANG_DEEPEP_NUM_MAX_DISPATCH_TOKENS_PER_RANK=188416
 export SGLANG_DISAGGREGATION_BOOTSTRAP_TIMEOUT=600
 export SGLANG_ENABLE_OVERLAP_PLAN_STREAM=1
-export SGLANG_NPU_FUSED_MOE_MODE=2
+export SGLANG_ENABLE_SPEC_V2=1
 export SGLANG_PREFILL_DELAYER_MAX_DELAY_PASSES=100
 export SGLANG_SCHEDULER_DECREASE_PREFILL_IDLE=1
 
@@ -186,7 +186,6 @@ python3 -m sglang.launch_server \
     --node-rank 0 \
     --attention-backend ascend \
     --device npu \
-    --quantization modelslim \
     --max-running-requests 432 \
     --context-length 8192 \
     --dtype bfloat16 \
@@ -196,6 +195,7 @@ python3 -m sglang.launch_server \
     --ep-dispatch-algorithm static \
     --disable-radix-cache \
     --moe-a2a-backend ascend_fuseep \
+    --fuseep-mode 2 \
     --speculative-algorithm EAGLE3 \
     --speculative-draft-model-path $DRAFT_MODEL_PATH \
     --speculative-num-steps 3 \