results_virtuoso_medium.py

"""Module containing performance results for the Virtuoso-Medium model."""

results_virtuoso_medium = {
    "name": "Virtuoso-Medium",
    "modelType": "Qwen2.5 32B",
    "configurations": [
        {
            "instanceType": "r8g.4xlarge",
            "quantization": "Q4_0_4_8",
            "container": "llama.cpp 11/27/24",
            "status": "OK",
            "tokensPerSecond": "10.5",
            "notes": "-fa",
        },
        {
            "instanceType": "g5.12xlarge",
            "quantization": "none",
            "container": "LMI 0.30+vLLM 0.6.2",
            "status": "OK",
            "tokensPerSecond": "23",
            "notes": '"OPTION_MAX_MODEL_LEN": "16384",\n"TENSOR_PARALLEL_DEGREE": "max",',
        },
        {
            "instanceType": "g6.12xlarge",
            "quantization": "none",
            "container": "LMI 0.30+vLLM 0.6.2",
            "status": "OK",
            "tokensPerSecond": "14",
            "notes": '"OPTION_MAX_MODEL_LEN": "16384",\n"TENSOR_PARALLEL_DEGREE": "max",',
        },
        {
            "instanceType": "g6e.12xlarge",
            "quantization": "none",
            "container": "vLLM0.6.4.post1",
            "status": "OK",
            "tokensPerSecond": "36",
            "notes": "--tensor-parallel-size 4 --max-model-len 16384",
        },
        {
            "instanceType": "g6e.12xlarge (2 GPUs)",
            "quantization": "none",
            "container": "vLLM 0.6.4.post1",
            "status": "OK",
            "tokensPerSecond": "21",
            "notes": "--tensor-parallel-size 2 --max-model-len 16384",
        },
        {
            "instanceType": "g6e.48xlarge",
            "quantization": "none",
            "container": "LMI 0.30+vLLM 0.6.2",
            "status": "OK",
            "tokensPerSecond": "48",
            "notes": "",
        },
        {
            "instanceType": "p4d.24xlarge",
            "quantization": "none",
            "container": "LMI 0.30+vLLM 0.6.2",
            "status": "OK",
            "tokensPerSecond": "72.5",
            "notes": '"OPTION_MAX_MODEL_LEN": "32768",\n"TENSOR_PARALLEL_DEGREE": "max",',
        },
        {
            "instanceType": "p5.48xlarge",
            "quantization": "none",
            "container": "vLLM 0.6.4.post1",
            "status": "OK",
            "tokensPerSecond": "117",
            "notes": "--tensor-parallel-size 8",
        },
        {
            "instanceType": "p5.48xlarge (4 GPUs)",
            "quantization": "none",
            "container": "vLLM 0.6.4.post1",
            "status": "OK",
            "tokensPerSecond": "88",
            "notes": "--tensor-parallel-size 4",
        },
        {
            "instanceType": "p5.48xlarge (2 GPUs)",
            "quantization": "none",
            "container": "vLLM 0.6.4.post1",
            "status": "OK",
            "tokensPerSecond": "58",
            "notes": "--tensor-parallel-size 2",
        },
        {
            "instanceType": "p5.48xlarge (1 GPU)",
            "quantization": "none",
            "container": "vLLM 0.6.4.post1",
            "status": "OK",
            "tokensPerSecond": "38",
            "notes": "--tensor-parallel-size 1",
        },
    ],
}