arise-predictions/config/example-demo-mlcommons-config.yaml at main · namasl/arise-predictions · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
                     # Demo Scenario 1 on Inference Throughput Data
# Given a fixed configuration and variable configuration, predict throughput
# for each combination. Use this to find optimal trade-offs.

estimators:
  - target_variable: "tokens_per_second"
    estimator_file: estimator-nonlinear-XGBoost-Regressor-tokens_per_second.pkl
    greater_is_better: True

# fixed configuration
fixed_values:
  - "Model MLC": "gptj-99"
  - "# of Nodes": 1
  - Processor: "AMD EPYC 9654"
  - Scenario: "Offline"

# variable configuration
variable_values:
  - Accelerator:
    - "AMD Instinct MI300X-NPS1-SPX-192GB-750W"
    - "AMD MI300X-NPS1-SPX-192GB-750W"
    - "NVIDIA H100-NVL-94GB"
    - "NVIDIA H100-PCIe-80GB"
    - "NVIDIA H100-SXM-80GB"
    - "NVIDIA L40S"
    - "NVIDIA GH200 Grace Hopper Superchip 96GB"
    - "NVIDIA GH200 Grace Hopper Superchip 144GB"
    - "NVIDIA H200-SXM-141GB-CTS"
    - "NVIDIA H200-SXM-141GB"
    - "NVIDIA B200-SXM-180GB"
  - "# of Accelerators":
    - 1
    - 4
    - 8
  - "Host Processor Core Count":
    - 32
    - 48
    - 64
    - 96
    - 120