Move all experiments under hardware-specific folders: - experiments/h200/ : H200 GPU experiments (15 dirs) - experiments/h20/ : H20 GPU experiments (2 dirs) - experiments/p800/ : Kunlun P800 experiments (3 dirs) - experiments/pro6000/ : RTX 6000D experiments (2 dirs) This improves discoverability and keeps hardware-specific configs isolated from each other.
183 lines
5.4 KiB
JSON
183 lines
5.4 KiB
JSON
{
|
|
"metadata": {
|
|
"experiment": "dsv4_h200_vllm",
|
|
"run_id": "20260708-062348",
|
|
"timestamp": "2026-07-08T06:23:48+00:00",
|
|
"model": "/data/models/DeepSeek-V4-Flash",
|
|
"backend": "vllm",
|
|
"engine": "vllm",
|
|
"hardware": "8x NVIDIA H200 143GB",
|
|
"accelerator": "NVIDIA H200",
|
|
"chip": "nvidia_h200",
|
|
"script": "experiments/dsv4_h200_vllm/run_bench.sh",
|
|
"env": "/data/user1/yy/envs/vllm",
|
|
"git_commit": "acf2e3d",
|
|
"git_dirty": "dirty",
|
|
"description": "H200 native vLLM baseline benchmark for DeepSeek-V4-Flash"
|
|
},
|
|
"config": {
|
|
"tp": 4,
|
|
"kv_cache_dtype": "fp8",
|
|
"block_size": 256,
|
|
"max_num_seqs": 256,
|
|
"port": 30005,
|
|
"num_prompts": 128,
|
|
"scenarios": [
|
|
"32 512 256",
|
|
"128 512 256",
|
|
"32 4000 512"
|
|
]
|
|
},
|
|
"scenarios": [
|
|
{
|
|
"name": "c128_i512_o256",
|
|
"config": {
|
|
"concurrency": 128,
|
|
"input_len": 512,
|
|
"output_len": 256,
|
|
"dataset": "random",
|
|
"num_prompts": 128
|
|
},
|
|
"metrics": {
|
|
"success": 128,
|
|
"failed": 0,
|
|
"duration_s": 5.98181911901338,
|
|
"request_throughput": 21.39817293925662,
|
|
"input_token_throughput": 5892.521873147793,
|
|
"output_token_throughput": 2825.2275208862256,
|
|
"total_token_throughput": 8717.749394034017,
|
|
"total_input_tokens": 35248,
|
|
"total_output_tokens": 16900,
|
|
"e2e_ms": {
|
|
"mean": 3786.2239771795885,
|
|
"p50": 3966.506554497755,
|
|
"p90": 5763.936419402307,
|
|
"p95": 5857.902326301701,
|
|
"p99": 5919.940241662553
|
|
},
|
|
"ttft_ms": {
|
|
"mean": 634.9285282819892,
|
|
"p50": 651.9191234983737,
|
|
"p90": 872.5053475049208,
|
|
"p95": 873.6197502956202,
|
|
"p99": 879.9700354097877
|
|
},
|
|
"tpot_ms": {
|
|
"mean": 28.086932902452045,
|
|
"p50": 24.940565250290092,
|
|
"p90": 33.6093050082046,
|
|
"p95": 40.66724980024446,
|
|
"p99": 91.16725345989823
|
|
},
|
|
"itl_ms": {
|
|
"mean": 24.055493866406856,
|
|
"p50": 22.871914501592983,
|
|
"p90": 26.806968497112393,
|
|
"p95": 27.607820050616283,
|
|
"p99": 32.99604622545303
|
|
}
|
|
},
|
|
"raw_file": "experiments/dsv4_h200_vllm/results/20260708-062348/raw_outputs/vllm_0708_128_512_256.jsonl"
|
|
},
|
|
{
|
|
"name": "c32_i4000_o512",
|
|
"config": {
|
|
"concurrency": 32,
|
|
"input_len": 4000,
|
|
"output_len": 512,
|
|
"dataset": "random",
|
|
"num_prompts": 128
|
|
},
|
|
"metrics": {
|
|
"success": 128,
|
|
"failed": 0,
|
|
"duration_s": 63.96937208699819,
|
|
"request_throughput": 2.000957580542143,
|
|
"input_token_throughput": 4229.977427822796,
|
|
"output_token_throughput": 505.9765156984964,
|
|
"total_token_throughput": 4735.953943521293,
|
|
"total_input_tokens": 270589,
|
|
"total_output_tokens": 32367,
|
|
"e2e_ms": {
|
|
"mean": 15397.358620969157,
|
|
"p50": 10554.476282501128,
|
|
"p90": 39473.02784240455,
|
|
"p95": 42952.8573812,
|
|
"p99": 46968.37465279941
|
|
},
|
|
"ttft_ms": {
|
|
"mean": 917.0502477351192,
|
|
"p50": 236.0078425044776,
|
|
"p90": 2404.053003701846,
|
|
"p95": 2853.309046203503,
|
|
"p99": 6638.214706212312
|
|
},
|
|
"tpot_ms": {
|
|
"mean": 62.674261138657926,
|
|
"p50": 38.76617610605361,
|
|
"p90": 141.09892115795085,
|
|
"p95": 165.57312805125648,
|
|
"p99": 219.71921567381835
|
|
},
|
|
"itl_ms": {
|
|
"mean": 57.50244012726164,
|
|
"p50": 14.738622994627804,
|
|
"p90": 19.66081059363205,
|
|
"p95": 142.28972980054095,
|
|
"p99": 1669.9934365210356
|
|
}
|
|
},
|
|
"raw_file": "experiments/dsv4_h200_vllm/results/20260708-062348/raw_outputs/vllm_0708_32_4000_512.jsonl"
|
|
},
|
|
{
|
|
"name": "c32_i512_o256",
|
|
"config": {
|
|
"concurrency": 32,
|
|
"input_len": 512,
|
|
"output_len": 256,
|
|
"dataset": "random",
|
|
"num_prompts": 128
|
|
},
|
|
"metrics": {
|
|
"success": 128,
|
|
"failed": 0,
|
|
"duration_s": 38.25599173500086,
|
|
"request_throughput": 3.3458811076355204,
|
|
"input_token_throughput": 921.3720100151314,
|
|
"output_token_throughput": 441.7608649925023,
|
|
"total_token_throughput": 1363.1328750076339,
|
|
"total_input_tokens": 35248,
|
|
"total_output_tokens": 16900,
|
|
"e2e_ms": {
|
|
"mean": 9197.956188436932,
|
|
"p50": 6626.5614014992025,
|
|
"p90": 20719.612499210045,
|
|
"p95": 27486.685669697905,
|
|
"p99": 28020.42960330684
|
|
},
|
|
"ttft_ms": {
|
|
"mean": 1228.4664978981255,
|
|
"p50": 132.46312249975745,
|
|
"p90": 3753.0612517031777,
|
|
"p95": 11441.480474699347,
|
|
"p99": 11446.530493390019
|
|
},
|
|
"tpot_ms": {
|
|
"mean": 79.49513794690715,
|
|
"p50": 35.04621934244705,
|
|
"p90": 158.92302877744768,
|
|
"p95": 246.69572046122963,
|
|
"p99": 674.493606643706
|
|
},
|
|
"itl_ms": {
|
|
"mean": 60.85736878373586,
|
|
"p50": 15.390933993330691,
|
|
"p90": 47.24743370315991,
|
|
"p95": 57.582202005869476,
|
|
"p99": 237.64244167061403
|
|
}
|
|
},
|
|
"raw_file": "experiments/dsv4_h200_vllm/results/20260708-062348/raw_outputs/vllm_0708_32_512_256.jsonl"
|
|
}
|
|
]
|
|
} |