-
Notifications
You must be signed in to change notification settings - Fork 82
/
Copy pathbenchmarks.yml
67 lines (61 loc) · 2.27 KB
/
benchmarks.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
---
common_options: &common_options
data:
throughput:
skip: 2
regexp: 'throughput: *(.*?) samples\/sec'
loss:
reduction_type: 'final'
regexp: 'loss: *(\d*\.\d*)'
output:
- [samples/sec, 'throughput']
- [loss, 'loss']
config_options: &config_options
requirements_path: requirements.txt
pre_run_commands: [make clean, make all]
pytorch_swin_tiny_train_real_pod16:
<<: [*common_options, *config_options]
description: PyTorch Sliding window transformer throughput benchmark using real data.
cmd: >-
python train_swin.py
--cfg SWIN_TINY
--data-path $DATASETS_DIR/imagenet-raw-dataset/
--checkpoint-output-dir ./output/swin_tiny_224/
--training-steps 10
pytorch_swin_base_384_train_real_pod16:
<<: [*common_options, *config_options]
description: PyTorch Sliding window transformer throughput benchmark using real data.
cmd: >-
python train_swin.py
--cfg SWIN_BASE_384
--data-path $DATASETS_DIR/imagenet-raw-dataset/
--checkpoint-output-dir ./output/swin_base_384/
--training-steps 10
pytorch_swin_base_linear_train_real_pod16:
<<: [*common_options, *config_options]
description: PyTorch Sliding window transformer throughput benchmark using real data.
cmd: >-
python train_swin.py
--cfg SWIN_BASE_LINEAR
--data-path $DATASETS_DIR/imagenet-raw-dataset/
--checkpoint-output-dir ./output/swin_base_linear/
--training-steps 10
pytorch_swin_large_finetune_real_pod16:
<<: [*common_options, *config_options]
description: PyTorch Sliding window transformer throughput benchmark using real data.
cmd: >-
python train_swin.py
--cfg SWIN_LARGE_224_22K_FINETUNE_1K_FP16_POD16
--data-path $DATASETS_DIR/imagenet-raw-dataset/
--checkpoint-output-dir ./output/swin_large/
--training-steps 10
pytorch_swin_large_finetune_real_pod16_conv:
<<: [*common_options, *config_options]
description: PyTorch Sliding window transformer throughput benchmark using real data.
cmd: >-
python train_swin.py
--cfg SWIN_LARGE_224_22K_FINETUNE_1K_FP16_POD16
--data-path $DATASETS_DIR/imagenet-raw-dataset/
--checkpoint-output-dir ./output/swin_large/
--pretrained-model $DATASETS_DIR/ckpts/swin_large_patch4_window7_224_22k.pth
--wandb