from chutes.chute import NodeSelector
from chutes.chute.template.vllm import build_vllm_chute

chute = build_vllm_chute(
    tee=True,
    username="chutes",
    model_name="Qwen/Qwen3-235B-A22B-Thinking-2507",
    revision="4cd68849b2a0fadb84866c703b133aa8c8636130",
    image="chutes/vllm:nightly-2026030502",
    concurrency=40,
    node_selector=NodeSelector(
        gpu_count=8,
        include=["h200", "b200", "b300"],
    ),
    engine_args=(
        "--enable-auto-tool-choice "
        "--tool-call-parser hermes "
        "--reasoning-parser deepseek_r1 "
        "--max-num-batched-tokens 4096 "
        "--max-completion-tokens 16384 "
        "--max-stream-completion-tokens 65536 "
        "--max-num-seqs 40 "
        "--max-cudagraph-capture-size 40 "
        "--gpu-memory-utilization 0.85"
    ),
)
chute.chute._name = "Qwen/Qwen3-235B-A22B-Thinking-2507-TEE"