engine = "vllm"
name = "GLM-5.1 NVFP8"
docker_image = "vllm/vllm-openai:glm51-cu130"
model = "lukealonso/GLM-5.1-NVFP4"
parent_model = "zai-org/GLM-5.1"

required_nodes = 4
distributed_backend = "mp"
pipeline_parallel_size = 0

args = [
    "--speculative-config.method", "mtp",
    "--speculative-config.num_speculative_tokens", "3",
    "--tool-call-parser", "glm47",
    "--reasoning-parser", "glm45",
    "--enable-auto-tool-choice",
    "--chat-template-content-format", "string",
    "--attention-backend", "TRITON_MLA",
    "--disable-flashinfer",
    "--disable-custom-all-reduce",
    "--enable-prefix-caching",
    "--enable-chunked-prefill",
    "--kv-cache-dtype", "fp8",
    "--max-model-len", "32768",
    "--max-num-batched-tokens", "6144",
    "--max-num-seqs", "48",
    "--gpu-memory-utilization", "0.88",
    "--trust-remote-code",
    "--served-model-name", "glm51", "default",
]

[env]
NCCL_DEBUG = "WARN"
NCCL_IB_MERGE_NICS = "1"
NCCL_MAX_NCHANNELS = "12"