kebab/config.yaml at main · LancerLab/kebab · 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
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
build:
  mode: release
  optimization: O3
  cuda_arch: auto
benchmark:
  warmup_runs: 20
  measurement_runs: 100
  batch_sizes:
  - 256
  - 512
  - 1024
  - 2048
  - 4096
  data_types:
  - float32
  - float16
profiling:
  ncu_metrics:
  - sm__throughput.avg.pct_of_peak_sustained_elapsed
  - dram__throughput.avg.pct_of_peak_sustained_elapsed
  - sm__sass_thread_inst_executed_op_fadd_pred_on.sum
  - sm__sass_thread_inst_executed_op_fmul_pred_on.sum
  - sm__sass_thread_inst_executed_op_ffma_pred_on.sum
  - smsp__sass_average_data_bytes_per_sector_mem_global_op_ld.pct
  - smsp__sass_average_data_bytes_per_sector_mem_global_op_st.pct
  - sm__inst_executed_pipe_tensor.sum
  sections:
  - SpeedOfLight
  - MemoryWorkloadAnalysis
  - ComputeWorkloadAnalysis
  - Occupancy
operators:
  elementwise_add:
    enabled: true
    gpu_id: 0
    sizes:
    - 256
    - 512
    - 1024
    - 2048
    - 4096
    impl: cuda
    init_method: rand-rand
  gemm:
    enabled: true
    gpu_id: 1
    impl:
    - cuda
    versions:
    - 2
    - 3
    - 4
    - 5
    - 10
    - 41
    init_method: rand-rand
    matrix_sizes:
    - 2048
    - 4096
    - 8192
    modes:
    - RC
    precisions:
    - float16
    tile_sizes:
    - 128
    - 128
    - 64
    verbose: false
  fft:
    enabled: false
    gpu_id: 0
    sizes:
    - 1024
    - 2048
    - 4096
    - 8192
  conv2d:
    enabled: false
    gpu_id: 0
    channels:
    - 32
    - 64
    - 128
    kernel_sizes:
    - 3
    - 5
    - 7
  reduction:
    enabled: false
    gpu_id: 0
    sizes:
    - 1024
    - 2048
    - 4096
    - 8192