Extract mamba state radix tree

[crystalxyz]

Start sglang server with Qwen3.5-9B model

python -m sglang.launch_server --model-path Qwen/Qwen3.5-9B --port 8000 --tp-size 1 --mem-fraction-static 0.8 --context-length 100000 --reasoning-parser qwen3 \
  --mamba-track-interval 256

Benchmark script

python -m sglang.bench_serving --backend sglang-oai-chat --dataset-name sharegpt --host 127.0.0.1 --port 8000 --num-prompts 10

Extract radix tree

curl -X POST http://localhost:8000/debug/dump_mamba_tree \
    -H "Content-Type: application/json" \
    -d '{"dump_dir": "tmp/mamba_states"}'

Run analysis script

python tmp/scripts/analyze_mamba_states.py tmp/mamba_states