from chutes.chute import NodeSelector
from chutes.chute.template.sglang import build_sglang_chute

chute = build_sglang_chute(
    tee=True,
    username="chutes",
    readme="Qwen/Qwen3.6-27B-FP8 with DFlash speculative decoding",
    model_name="Qwen/Qwen3.6-27B-FP8",
    name="Qwen/Qwen3.6-27B-TEE",
    image="chutes/sglang:0.5.17-qwen38",
    official_package=True,
    concurrency=24,
    revision="e89b16ebf1988b3d6befa7de50abc2d76f26eb09",
    node_selector=NodeSelector(
        gpu_count=1,
        include=["pro_6000"],
    ),
    engine_args=(
        "--trust-remote-code "
        "--reasoning-parser qwen3 "
        "--tool-call-parser qwen3_coder "
        "--mamba-scheduler-strategy extra_buffer "
        "--mem-fraction-static 0.87 "
        "--max-mamba-cache-size 48 "
        "--cuda-graph-max-bs 24 "
        "--disable-custom-all-reduce"
    ),
)