59799a3687
Co-authored-by: zhouhaizhu.zhz <zhouhaizhu.zhz@alibaba-inc.com> Co-authored-by: LinSiyuan814 <linsiyuan.lsy@alibaba-inc.com> Co-authored-by: hzh0425 <hzh0425@apache.org>
23 lines
503 B
JSON
23 lines
503 B
JSON
{
|
|
"platform": {
|
|
"accelerator": {"name": "a100_sxm", "hbm_capacity_gb": 80},
|
|
"disk_read_bandwidth_gb": 8,
|
|
"disk_write_bandwidth_gb": 8,
|
|
"memory_read_bandwidth_gb": 64,
|
|
"memory_write_bandwidth_gb": 64,
|
|
"num_device_per_node": 8
|
|
},
|
|
"predictor": {
|
|
"name": "ml",
|
|
"database_path": "../assets/model.pkl",
|
|
"latency_scale": 1.0
|
|
},
|
|
"scheduler": {
|
|
"tp_size": 1,
|
|
"ep_size": 1,
|
|
"dp_size": 1,
|
|
"backend_name": "sglang",
|
|
"backend_version": "0.5.9"
|
|
}
|
|
}
|