[NPU] Add mxfp4-w4a8 MOE Quantization Support for NPU (#30318)

docs/docs/advanced_features/quantization.mdx

@@ -61,7 +61,7 @@ The following table summarizes quantization method support across NVIDIA and AMD
       <td>No</td>
       <td>No</td>
       <td>Yes (A5)</td>
-      <td>Ascend NPU only; online W4A8 for Qwen3 dense LLM (MXFP4 weights + MXFP8 activations) on A5 series; offline <code>W4A8_MXFP</code> checkpoints are auto-detected via <code>modelslim</code></td>
+      <td>Ascend NPU only; online W4A8 for Qwen3 dense LLM (MXFP4 weights + MXFP8 activations) on A5 series; offline <code>W4A8_MXFP</code> dense and MoE checkpoints are auto-detected via <code>modelslim</code></td>
     </tr>
     <tr>
       <td><code>blockwise_int8</code></td>
@@ -808,6 +808,7 @@ MindStudio-ModelSlim (msModelSlim) is a model offline quantization compression t
     - [x]  ```W8A8_DYNAMIC``` linear with online quantization of activations
     - [x]  ```W4A4_DYNAMIC``` MOE with online quantization of activations
     - [x]  ```W4A8_DYNAMIC``` MOE with online quantization of activations
+    - [x]  ```W4A8_MXFP``` MOE with dynamic MXFP8 activation quantization
     - [x]  ```W8A8_DYNAMIC``` MOE with online quantization of activations
     - [ ]  ```W4A8``` linear TBD
     - [ ]  ```W4A16``` linear TBD

docs/docs/hardware-platforms/ascend-npus/optimization/quantization.mdx

@@ -68,6 +68,14 @@ SGLang supports **mix-bits** quantization (independently defines and loads each
       <td><strong style={{color: 'green'}}>√</strong></td>
       <td><strong style={{color: 'red'}}>x</strong></td>
     </tr>
+    <tr>
+      <td><a href="https://github.com/sgl-project/sglang/pull/30318">MXFP4 W4A8 (ModelSlim)</a></td>
+      <td>MoE</td>
+      <td><strong style={{color: 'red'}}>x</strong></td>
+      <td><strong style={{color: 'red'}}>x</strong></td>
+      <td><strong style={{color: 'green'}}>√</strong></td>
+      <td><strong style={{color: 'red'}}>x</strong></td>
+    </tr>
     <tr>
       <td><a href="https://github.com/sgl-project/sglang/pull/23795">MXFP4 W4A4</a></td>
       <td>Linear</td>
@@ -418,6 +426,23 @@ python3 -m sglang.launch_server \
 > - The packed-FP4 dtype passed to the NPU ops (`dst_type` / `x2_dtype` / `input_dtype`) must be resolved from `torch_npu.float4_e2m1fn_x2` (an int enum), not the `torch.float4_e2m1fn_x2` dtype object, which recent op-plugin builds reject.
 > - Online and offline share the same kernel path and layout; they differ only in the weight source (RTN at load vs msmodelslim calibration).
 
+**ModelSlim W4A8 MXFP4 for LLM MoE models:**
+
+SGLang auto-detects offline ModelSlim `W4A8_MXFP` MoE checkpoints from `quant_model_description.json`; do not pass `--quantization`. This path requires Ascend A5 or newer.
+
+```bash Command
+MODEL_PATH=/path/to/w4a8-mxfp4-moe-model
+python3 -m sglang.launch_server \
+    --model-path "$MODEL_PATH" \
+    --device npu --attention-backend ascend \
+    --host 0.0.0.0 --port 30000 \
+    --tp-size 1
+```
+
+> **Implementation Notes:**
+> - ModelSlim supplies packed MXFP4 `w13` and `w2` expert weights with UE8M0 block scales (block size 32).
+> - Ascend TP and DeepEP dispatch activations as BF16; this path does not request MXFP8 dispatch. SGLang dynamically quantizes each expert input to MXFP8 immediately before grouped matmul.
+
 **MXFP4 W4A4 for LLM dense models (e.g. Qwen3 / Qwen3.5):**
 
 LLM dense W4A4 (MXFP4 4-bit weights + 4-bit activations) Linear support was added in [PR #23795](https://github.com/sgl-project/sglang/pull/23795). Requires Ascend A5 series (Ascend 950) or newer — the dual-level online path uses the `DualLevelQuantBatchMatmul` op, which A2/A3 lack. On the Ascend NPU backend `--quantization mxfp4` selects this W4A4 path (on GPU the same flag selects the upstream OCP MXFP4 MoE config instead).