default.yaml 4.48 KB
Newer Older
1
# SuperBench Config
2
version: v0.4
3
superbench:
4
  enable: null
5
  monitor:
6
    enable: true
7
8
    sample_duration: 1
    sample_interval: 10
9
10
  var:
    default_local_mode: &default_local_mode
11
      enable: true
12
13
14
15
      modes:
        - name: local
          proc_num: 8
          prefix: CUDA_VISIBLE_DEVICES={proc_rank}
16
17
          parallel: yes
    default_pytorch_mode: &default_pytorch_mode
18
19
20
21
      enable: true
      modes:
        - name: torch.distributed
          proc_num: 8
22
          node_num: 1
23
24
      frameworks:
        - pytorch
25
26
27
28
    common_model_config: &common_model_config
      duration: 0
      num_warmup: 16
      num_steps: 128
29
      batch_size: 1
30
31
32
33
34
35
      precision:
        - float32
        - float16
      model_action:
        - train
  benchmarks:
36
37
38
39
    nccl-bw:
      enable: true
      modes:
        - name: local
40
41
42
43
          proc_num: 1
          parallel: no
      parameters:
        ngpus: 8
44
45
46
47
48
49
50
51
52
53
54
    ib-loopback:
      enable: true
      modes:
        - name: local
          proc_num: 4
          prefix: PROC_RANK={proc_rank} IB_DEVICES=0,2,4,6 NUMA_NODES=1,0,3,2
          parallel: yes
        - name: local
          proc_num: 4
          prefix: PROC_RANK={proc_rank} IB_DEVICES=1,3,5,7 NUMA_NODES=1,0,3,2
          parallel: yes
55
56
57
    disk-benchmark:
      enable: false
      modes:
58
59
        - name: local
          proc_num: 1
60
61
62
63
          parallel: no
      parameters:
        block_devices:
          - /dev/nvme0n1
64
65
66
67
68
69
70
71
72
73
74
    cpu-memory-bw-latency:
      enable: false
      modes:
        - name: local
          proc_num: 1
          parallel: no
      parameters:
        tests:
          - bandwidth_matrix
          - latency_matrix
          - max_bandwidth
75
76
77
78
79
    mem-bw:
      enable: true
      modes:
        - name: local
          proc_num: 8
80
          prefix: CUDA_VISIBLE_DEVICES={proc_rank} numactl -N $(({proc_rank}/2))
81
          parallel: no
82
    gpu-copy-bw:
83
      enable: true
84
85
86
87
      modes:
        - name: local
          parallel: no
      parameters:
88
89
        mem_type:
          - htod
90
91
92
93
94
          - dtoh
          - dtod
        copy_type:
          - sm
          - dma
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
    kernel-launch:
      <<: *default_local_mode
    gemm-flops:
      <<: *default_local_mode
    cudnn-function:
      <<: *default_local_mode
    cublas-function:
      <<: *default_local_mode
    matmul:
      <<: *default_local_mode
      frameworks:
        - pytorch
    sharding-matmul:
      <<: *default_pytorch_mode
    computation-communication-overlap:
      <<: *default_pytorch_mode
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
    ib-traffic:
      enable: false
      modes:
        - name: mpi
          proc_num: 1
    gpcnet-network-test:
      enable: false
      modes:
        - name: mpi
          proc_num: 1
          mca:
            pml: ucx
            btl: ^uct
            btl_tcp_if_include: eth0
          env:
            UCX_NET_DEVICES: mlx5_0:1
    gpcnet-network-load-test:
      enable: false
      modes:
        - name: mpi
          proc_num: 1
          mca:
            pml: ucx
            btl: ^uct
            btl_tcp_if_include: eth0
          env:
            UCX_NET_DEVICES: mlx5_0:1
    tcp-connectivity:
      enable: false
      modes:
        - name: local
          parallel: no
      parameters:
        port: 22
145
146
    ort-inference:
      <<: *default_local_mode
147
148
      parameters:
        batch_size: 1
149
150
151
152
153
154
155
156
157
158
159
160
    tensorrt-inference:
      <<: *default_local_mode
      parameters:
        pytorch_models:
          - resnet50
          - resnet101
          - resnet152
          - densenet169
          - densenet201
          - bert-base
          - bert-large
        seq_length: 224
161
        batch_size: 1
162
        precision: int8
163
164
    gpt_models:
      <<: *default_pytorch_mode
165
166
167
      models:
        - gpt2-small
        - gpt2-large
168
      parameters:
169
        <<: *common_model_config
170
    bert_models:
171
      <<: *default_pytorch_mode
172
173
174
175
      models:
        - bert-base
        - bert-large
      parameters:
176
        <<: *common_model_config
177
    lstm_models:
178
      <<: *default_pytorch_mode
179
180
181
      models:
        - lstm
      parameters:
182
183
184
        <<: *common_model_config
    resnet_models:
      <<: *default_pytorch_mode
185
186
187
188
      models:
        - resnet50
        - resnet101
        - resnet152
189
190
191
192
193
      parameters:
        <<: *common_model_config
    densenet_models:
      <<: *default_pytorch_mode
      models:
194
195
        - densenet169
        - densenet201
196
197
198
199
200
      parameters:
        <<: *common_model_config
    vgg_models:
      <<: *default_pytorch_mode
      models:
201
202
203
204
205
        - vgg11
        - vgg13
        - vgg16
        - vgg19
      parameters:
206
        <<: *common_model_config