[Fix] Route asymmetric-KV models to fa4 on SM100 and pin MiMoV2 FP8 MoE to flashinfer_trtllm (#32818)

docs_new/cookbook/autoregressive/Xiaomi/MiMo-V2.5.mdx

@@ -79,7 +79,9 @@ import { MiMoV25Deployment } from '/src/snippets/autoregressive/mimo-v25-deploym
 **MiMo-V2.5 (310B):**
 - The checkpoint has a TP=4-interleaved fused `qkv_proj`; attention-TP per DP group **must** be 4. Use `--dp = TP / 4`; for TP > 4 this also requires DP-attention. Total GPUs must be a multiple of 4. A bare `--tp 8` without `--dp 2` will fail to load with `MiMoV2 fused qkv_proj checkpoint is TP=4-interleaved; got attention tp_size=8`.
 - Single-node deployments: H100/H200 8× GPUs (`--tp 8 --dp 2`), B200 4× GPUs (`--tp 4`, dp=1, no DP-attn flag needed), GB300 4× GPUs (`--tp 4`, single NVL4 node). FP8 quantization.
+- On Blackwell, pass `--attention-backend fa4`: MiMoV2's asymmetric KV (`head_dim` 192 / `v_head_dim` 128) fails on the SM100 default `trtllm_mha`, which requires equal K/V widths.
 - On Blackwell, pass `--mm-attention-backend fa4` for the V2.5 vision encoder. The checkpoint config requests FlashAttention-3 internally, but SGLang rejects FA3 on Blackwell and expects FA4 for multimodal attention.
+- On Blackwell, pass `--moe-runner-backend flashinfer_trtllm`; the default `auto` falls through to the triton fused-MoE runner, ~12% slower at bs=1 decode.
 - `--enable-dp-lm-head` and `--mm-enable-dp-encoder` are required whenever `--enable-dp-attention` is on, to keep LM head and encoder sharding consistent.
 - EAGLE MTP uses the checkpoint's MTP weights. Enable with `--speculative-algorithm EAGLE` and `--enable-multi-layer-eagle` (both Hopper and Blackwell).
 - **Multimodal**: Supports image, video, and audio understanding; see Section 4.3 for invocation examples.

docs_new/src/snippets/autoregressive/mimo-v25-deployment.jsx

@@ -321,9 +321,9 @@ export const MiMoV25Deployment = () => {
 
     if (multinode) flags.push(...multiNodeFlags(nnodes));
 
-    // MoE backend: Blackwell uses flashinfer_trtllm (hardware-driven); Hopper
+    // MoE backend: Blackwell uses flashinfer_trtllm (both variants); Hopper
     // optionally uses DeepEP (toggle).
-    if (isPro && blackwell) {
+    if (blackwell) {
       flags.push("  --moe-runner-backend flashinfer_trtllm");
     } else if (useDeepep) {
       flags.push("  --moe-a2a-backend deepep");
@@ -349,7 +349,11 @@ export const MiMoV25Deployment = () => {
         flags.push(`  --model-loader-extra-config '{"enable_multithread_load": true, "num_threads": 64}'`);
       }
     } else {
-      if (blackwell) flags.push("  --mm-attention-backend fa4");
+      if (blackwell) {
+        // fa4 is required, not tuning: trtllm_mha rejects MiMoV2's 192/128 KV.
+        flags.push("  --attention-backend fa4");
+        flags.push("  --mm-attention-backend fa4");
+      }
       flags.push("  --mem-fraction-static 0.65");
       flags.push("  --chunked-prefill-size 16384");
     }