forked from pytorch/pytorch
-
Notifications
You must be signed in to change notification settings - Fork 0
135 lines (127 loc) · 8.03 KB
/
inductor-perf-test-nightly-aarch64.yml
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
97
98
99
100
101
102
103
104
105
106
107
108
109
110
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
name: inductor-perf-nightly-aarch64
on:
schedule:
# - cron: 0 7 * * 1-6
# - cron: 0 7 * * 0
# Does not perform max_autotune on CPU, so skip the weekly run setup
# Run 6 times everyday to see if perf instablity can be reproduced
# Will change this back
- cron: 0 */4 * * *
# NB: GitHub has an upper limit of 10 inputs here
workflow_dispatch:
inputs:
training:
# CPU for training is not typical, but leave the option open here
description: Run training (off by default)?
required: false
type: boolean
default: false
inference:
description: Run inference (on by default)?
required: false
type: boolean
default: true
default:
description: Run inductor_default?
required: false
type: boolean
default: true
dynamic:
description: Run inductor_dynamic_shapes?
required: false
type: boolean
default: false
aotinductor:
description: Run aot_inductor for inference?
required: false
type: boolean
default: false
benchmark_configs:
description: The list of configs used the benchmark
required: false
type: string
default: inductor_huggingface_perf_cpu_aarch64,inductor_timm_perf_cpu_aarch64,inductor_torchbench_perf_cpu_aarch64
concurrency:
group: ${{ github.workflow }}-${{ github.event.pull_request.number || github.ref_name }}-${{ github.ref_type == 'branch' && github.sha }}-${{ github.event_name == 'workflow_dispatch' }}-${{ github.event_name == 'schedule' }}
cancel-in-progress: true
permissions: read-all
jobs:
linux-jammy-aarch64-py3_10-inductor-build:
name: linux-jammy-aarch64-py3.10-inductor
uses: ./.github/workflows/_linux-build.yml
with:
runner: linux.arm64.m7g.4xlarge
build-environment: linux-jammy-aarch64-py3.10
docker-image-name: pytorch-linux-jammy-aarch64-py3.10-gcc11-inductor-benchmarks
test-matrix: |
{ include: [
{ config: "inductor_huggingface_perf_cpu_aarch64", shard: 1, num_shards: 9, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_huggingface_perf_cpu_aarch64", shard: 2, num_shards: 9, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_huggingface_perf_cpu_aarch64", shard: 3, num_shards: 9, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_huggingface_perf_cpu_aarch64", shard: 4, num_shards: 9, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_huggingface_perf_cpu_aarch64", shard: 5, num_shards: 9, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_huggingface_perf_cpu_aarch64", shard: 6, num_shards: 9, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_huggingface_perf_cpu_aarch64", shard: 7, num_shards: 9, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_huggingface_perf_cpu_aarch64", shard: 8, num_shards: 9, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_huggingface_perf_cpu_aarch64", shard: 9, num_shards: 9, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_timm_perf_cpu_aarch64", shard: 1, num_shards: 15, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_timm_perf_cpu_aarch64", shard: 2, num_shards: 15, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_timm_perf_cpu_aarch64", shard: 3, num_shards: 15, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_timm_perf_cpu_aarch64", shard: 4, num_shards: 15, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_timm_perf_cpu_aarch64", shard: 5, num_shards: 15, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_timm_perf_cpu_aarch64", shard: 6, num_shards: 15, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_timm_perf_cpu_aarch64", shard: 7, num_shards: 15, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_timm_perf_cpu_aarch64", shard: 8, num_shards: 15, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_timm_perf_cpu_aarch64", shard: 9, num_shards: 15, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_timm_perf_cpu_aarch64", shard: 10, num_shards: 15, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_timm_perf_cpu_aarch64", shard: 11, num_shards: 15, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_timm_perf_cpu_aarch64", shard: 12, num_shards: 15, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_timm_perf_cpu_aarch64", shard: 13, num_shards: 15, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_timm_perf_cpu_aarch64", shard: 14, num_shards: 15, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_timm_perf_cpu_aarch64", shard: 15, num_shards: 15, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_torchbench_perf_cpu_aarch64", shard: 1, num_shards: 12, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_torchbench_perf_cpu_aarch64", shard: 2, num_shards: 12, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_torchbench_perf_cpu_aarch64", shard: 3, num_shards: 12, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_torchbench_perf_cpu_aarch64", shard: 4, num_shards: 12, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_torchbench_perf_cpu_aarch64", shard: 5, num_shards: 12, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_torchbench_perf_cpu_aarch64", shard: 6, num_shards: 12, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_torchbench_perf_cpu_aarch64", shard: 7, num_shards: 12, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_torchbench_perf_cpu_aarch64", shard: 8, num_shards: 12, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_torchbench_perf_cpu_aarch64", shard: 9, num_shards: 12, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_torchbench_perf_cpu_aarch64", shard: 10, num_shards: 12, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_torchbench_perf_cpu_aarch64", shard: 11, num_shards: 12, runner: "linux.arm64.m7g.metal" },
{ config: "inductor_torchbench_perf_cpu_aarch64", shard: 12, num_shards: 12, runner: "linux.arm64.m7g.metal" },
]}
selected-test-configs: ${{ inputs.benchmark_configs }}
secrets:
HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
linux-jammy-aarch64-py3_10-inductor-test-nightly:
name: linux-jammy-aarch64-py3.10-inductor
uses: ./.github/workflows/_linux-test.yml
needs: linux-jammy-aarch64-py3_10-inductor-build
if: github.event.schedule == '0 */4 * * *'
with:
build-environment: linux-jammy-aarch64-py3.10
# Turn off dynamic-shapes and aotinductor tests for now, to have faster iteration for debugging perf instability.
# Will change this back
dashboard-tag: training-false-inference-true-default-true-dynamic-false-aotinductor-false
docker-image: ${{ needs.linux-jammy-aarch64-py3_10-inductor-build.outputs.docker-image }}
test-matrix: ${{ needs.linux-jammy-aarch64-py3_10-inductor-build.outputs.test-matrix }}
use-gha: anything-non-empty-to-use-gha
timeout-minutes: 720
secrets:
HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}
linux-jammy-aarch64-py3_10-inductor-test:
name: linux-jammy-aarch64-py3.10-inductor
uses: ./.github/workflows/_linux-test.yml
needs: linux-jammy-aarch64-py3_10-inductor-build
if: github.event_name == 'workflow_dispatch'
with:
build-environment: linux-jammy-aarch64-py3.10
dashboard-tag: training-${{ inputs.training }}-inference-${{ inputs.inference }}-default-${{ inputs.default }}-dynamic-${{ inputs.dynamic }}-aotinductor-${{ inputs.aotinductor }}
docker-image: ${{ needs.linux-jammy-aarch64-py3_10-inductor-build.outputs.docker-image }}
test-matrix: ${{ needs.linux-jammy-aarch64-py3_10-inductor-build.outputs.test-matrix }}
use-gha: anything-non-empty-to-use-gha
timeout-minutes: 720
secrets:
HUGGING_FACE_HUB_TOKEN: ${{ secrets.HUGGING_FACE_HUB_TOKEN }}