[Feature] Xiaomi MiMo-V2.6/MiMo-V2.6-Pro day0 support

[ollybbmonster]

Motivation

Day0 support for MiMo-V2.6/MiMo-V2.6-Pro in SGLang.

• mxfp4 MoE
• bf16 MoE router
• DFlash fo spec decoding

Launch command examples

MiMo V2.6 Flash model weights: https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL
MiMo V2.6 Pro model weights: https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL

MiMo V2 flash:

python3 -m sglang.launch_server \
              --model-path /model/MiMo-V2.5-FP4-DFlash/ \
              --served-model-name mimo-v2.5-fp4 \
              --log-level-http warning \
              --enable-cache-report \
              --pp-size 1 \
              --dp-size 1 \
              --tp-size 4 \
              --ep-size 4 \
              --moe-runner-backend deep_gemm \
              --moe-a2a-backend deepep \
              --deepep-mode auto \
              --page-size 1 \
              --host 0.0.0.0 \
              --port 9001 \
              --cuda-graph-max-bs-decode 64 \
              --max-running-requests 64 \
              --trust-remote-code \
              --mem-fraction-static 0.6 \
              --swa-full-tokens-ratio 0.03 \
              --chunked-prefill-size 49152 \
              --max-prefill-tokens 65536 \
              --reasoning-parser mimo \
              --tool-call-parser mimo \
              --attention-backend fa4 \
              --skip-server-warmup \
              --moe-dense-tp-size 1 \
              --enable-dp-lm-head \
              --speculative-algorithm DFLASH \
              --speculative-draft-model-path /model/MiMo-V2.5-FP4-DFlash/dflash/ \
              --speculative-num-draft-tokens 8 \
              --context-length 1048576 \
              --cuda-graph-backend-prefill=disabled \
              --disable-radix-cache \
              --mm-enable-dp-encoder \
              --mm-attention-backend fa4

MiMo V2 Pro:

python3 -m sglang.launch_server \
              --model-path /model/MiMo-V2.5-Pro-FP4-DFlash/ \
              --served-model-name mimo-v2.5-fp4 \
              --log-level-http warning \
              --enable-cache-report \
              --pp-size 1 \
              --dp-size 1 \
              --tp-size 8 \
              --ep-size 8 \
              --moe-runner-backend deep_gemm \
              --moe-a2a-backend deepep \
              --deepep-mode auto \
              --page-size 1 \
              --host 0.0.0.0 \
              --port 9001 \
              --cuda-graph-max-bs-decode 64 \
              --max-running-requests 64 \
              --trust-remote-code \
              --mem-fraction-static 0.7 \
              --swa-full-tokens-ratio 0.08 \
              --chunked-prefill-size 49152 \
              --max-prefill-tokens 65536 \
              --reasoning-parser mimo \
              --tool-call-parser mimo \
              --attention-backend fa4 \
              --skip-server-warmup \
              --moe-dense-tp-size 1 \
              --enable-dp-lm-head \
              --speculative-algorithm DFLASH \
              --speculative-draft-model-path /model/MiMo-V2.5-Pro-FP4-DFlash/dflash/ \
              --speculative-num-draft-tokens 8 \
              --context-length 1048576 \
              --cuda-graph-backend-prefill=disabled \
              --disable-radix-cache \
              --mm-enable-dp-encoder \
              --mm-attention-backend fa4

Tests

• pre-commit run --all-files passed.
• python -m py_compile of the changed files passed.
• Added registered GSM8K + accept-length coverage for FP4 MiMo-V2 DFlash: test/registered/e2e/speculative/test_dflash_mimo_v2_fp4.py::TestMiMoV2FP4DFlash::test_gsm8k.
  It needs a 8-GPU runner, passed.

---

CI States

Latest PR Test (Base): ✅ Run #35525488953
Latest PR Test (Extra): ❌ Run #35525488807
Latest PR Test (AMD ROCm 10): ❌ Run #35525488899

[ispobock]

/rerun-failed-ci base extra