[fix] resolve negative dimension when creating tensor in multi-layer MTP

[fromck]

Motivation

When I launch the MiMo-V2-Flash model on an H20, it has a problem creating a tensor with a negative dimension. The server arguments and the problem are as follows.

export MODEL_PATH=/ssd3/models/MiMo-V2-Flash
export PORT=8889
export HOST_IP=0.0.0.0

SGLANG_ENABLE_SPEC_V2=1 python3 -u -m sglang.launch_server \
    --model-path $MODEL_PATH \
    --served-model-name MiMo-V2-Flash \
    --page-size 64 \
    --max-total-tokens 798720    \
    --decode-log-interval 60 \
    --disable-radix-cache \
    --host $HOST_IP \
    --port $PORT \
    --trust-remote-code \
    --tp-size 8 \
     --dp-size 2 --enable-dp-attention \
    --max-running-requests 64 \
    --chunked-prefill-size 4096 \
    --disable-overlap-schedule \
    --attention-backend fa3 \
    --cuda-graph-max-bs 64 \
    --mem-fraction-static 0.9  --disable-cuda-graph \
    --init-expert-location trivial \
    --ep-num-redundant-experts 0 \
    --eplb-algorithm deepseek \
    --ep-dispatch-algorithm static \
    --ep-size 8 --speculative-algorithm EAGLE --speculative-num-steps 3 \
    --speculative-eagle-topk 1 \
    --speculative-num-draft-tokens 4

Modifications

Because spec_info.num_tokens_per_req has a default value of -1, we need to set a correct value.

Accuracy Tests

Speed Tests and Profiling

Checklist

• Format your code according to the Format code with pre-commit.
• Add unit tests according to the Run and add unit tests.
• Update documentation according to Write documentations.
• Provide accuracy and speed benchmark results according to Test the accuracy and Benchmark the speed.
• Follow the SGLang code style guidance.

Review and Merge Process

1. Ping Merge Oncalls to start the process. See the PR Merge Process.
2. Get approvals from CODEOWNERS and other reviewers.
3. Trigger CI tests with comments or contact authorized users to do so.
   • Common commands include /tag-and-rerun-ci, /tag-run-ci-label, /rerun-failed-ci
4. After green CI and required approvals, ask Merge Oncalls or people with Write permission to merge the PR.

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!