Skip to content

Commit cefc16b

Browse files
jantonguiraojantonguirao
committed
Code review fixes
Signed-off-by: Joaquin Anton Guirao <janton@nvidia.com>
1 parent 2a1ed8f commit cefc16b

File tree

2 files changed

+44
-30
lines changed

2 files changed

+44
-30
lines changed

‎docs/examples/use_cases/pytorch/resnet50/main.py

+5-4
Original file line numberDiff line numberDiff line change
@@ -93,12 +93,13 @@ def parse():
9393
'"dali" for DALI data loader, or "dali_proxy" for PyTorch dataloader with DALI proxy preprocessing.')
9494
parser.add_argument('--prof', default=-1, type=int,
9595
help='Only run 10 iterations for profiling.')
96-
parser.add_argument('--deterministic', action='store_true')
97-
96+
parser.add_argument('--deterministic',
97+
help="If enabled, random seeds are fixed to ensure deterministic results for reproducibility.",
98+
action='store_true')
9899
parser.add_argument('--fp16-mode', default=False, action='store_true',
99100
help='Enable half precision mode.')
100-
parser.add_argument('--loss-scale', type=float, default=1)
101-
parser.add_argument('--channels-last', type=bool, default=False)
101+
parser.add_argument('--loss-scale', type=float, help="Loss scaling factor for mixed precision training. Default is 1.", default=1)
102+
parser.add_argument('--channels-last', type=bool, help="Use channels-last memory format for model and data. Default is False.", default=False)
102103
parser.add_argument('-t', '--test', action='store_true',
103104
help='Launch test mode with preset arguments')
104105
args = parser.parse_args()

‎docs/examples/use_cases/pytorch/resnet50/pytorch-resnet50.rst

+39-26
Original file line numberDiff line numberDiff line change
@@ -54,29 +54,42 @@ Usage
5454
PyTorch ImageNet Training
5555
5656
positional arguments:
57-
DIR path(s) to dataset (if one path is provided, it is assumed to have subdirectories named "train" and "val"; alternatively, train and val paths can be specified directly by providing both paths as arguments)
58-
59-
optional arguments (for the full list please check `Apex ImageNet example
60-
<https://github.com/NVIDIA/apex/tree/master/examples/imagenet>`_)
61-
-h, --help show this help message and exit
62-
--arch ARCH, -a ARCH model architecture: alexnet | resnet | resnet101
63-
| resnet152 | resnet18 | resnet34 | resnet50 | vgg
64-
| vgg11 | vgg11_bn | vgg13 | vgg13_bn | vgg16
65-
| vgg16_bn | vgg19 | vgg19_bn (default: resnet18)
66-
-j N, --workers N number of data loading workers (default: 4)
67-
--epochs N number of total epochs to run
68-
--start-epoch N manual epoch number (useful on restarts)
69-
-b N, --batch-size N mini-batch size (default: 256)
70-
--lr LR, --learning-rate LR initial learning rate
71-
--momentum M momentum
72-
--weight-decay W, --wd W weight decay (default: 1e-4)
73-
--print-freq N, -p N print frequency (default: 10)
74-
--resume PATH path to latest checkpoint (default: none)
75-
-e, --evaluate evaluate model on validation set
76-
--pretrained use pre-trained model
77-
--dali_cpu use CPU based pipeline for DALI, for heavy GPU
78-
networks it may work better, for IO bottlenecked
79-
one like RN18 GPU default should be faster
80-
--data_loader Select data loader: "pytorch" for native PyTorch data loader,
81-
"dali" for DALI data loader, or "dali_proxy" for PyTorch dataloader with DALI proxy preprocessing.
82-
--fp16-mode enables mixed precision mode
57+
DIR path(s) to dataset (if one path is provided, it is assumed to have subdirectories named "train" and "val"; alternatively, train and val paths can be specified
58+
directly by providing both paths as arguments)
59+
60+
options:
61+
-h, --help show this help message and exit
62+
--arch ARCH, -a ARCH model architecture: alexnet | convnext_base | convnext_large | convnext_small | convnext_tiny | densenet121 | densenet161 | densenet169 | densenet201 |
63+
efficientnet_b0 | efficientnet_b1 | efficientnet_b2 | efficientnet_b3 | efficientnet_b4 | efficientnet_b5 | efficientnet_b6 | efficientnet_b7 | efficientnet_v2_l |
64+
efficientnet_v2_m | efficientnet_v2_s | get_model | get_model_builder | get_model_weights | get_weight | googlenet | inception_v3 | list_models | maxvit_t |
65+
mnasnet0_5 | mnasnet0_75 | mnasnet1_0 | mnasnet1_3 | mobilenet_v2 | mobilenet_v3_large | mobilenet_v3_small | regnet_x_16gf | regnet_x_1_6gf | regnet_x_32gf |
66+
regnet_x_3_2gf | regnet_x_400mf | regnet_x_800mf | regnet_x_8gf | regnet_y_128gf | regnet_y_16gf | regnet_y_1_6gf | regnet_y_32gf | regnet_y_3_2gf | regnet_y_400mf
67+
| regnet_y_800mf | regnet_y_8gf | resnet101 | resnet152 | resnet18 | resnet34 | resnet50 | resnext101_32x8d | resnext101_64x4d | resnext50_32x4d |
68+
shufflenet_v2_x0_5 | shufflenet_v2_x1_0 | shufflenet_v2_x1_5 | shufflenet_v2_x2_0 | squeezenet1_0 | squeezenet1_1 | swin_b | swin_s | swin_t | swin_v2_b | swin_v2_s
69+
| swin_v2_t | vgg11 | vgg11_bn | vgg13 | vgg13_bn | vgg16 | vgg16_bn | vgg19 | vgg19_bn | vit_b_16 | vit_b_32 | vit_h_14 | vit_l_16 | vit_l_32 | wide_resnet101_2 |
70+
wide_resnet50_2 (default: resnet18)
71+
-j N, --workers N number of data loading workers (default: 4)
72+
--epochs N number of total epochs to run
73+
--start-epoch N manual epoch number (useful on restarts)
74+
-b N, --batch-size N mini-batch size per process (default: 256)
75+
--lr LR, --learning-rate LR
76+
Initial learning rate. Will be scaled by <global batch size>/256: args.lr = args.lr*float(args.batch_size*args.world_size)/256. A warmup schedule will also be
77+
applied over the first 5 epochs.
78+
--momentum M momentum
79+
--weight-decay W, --wd W
80+
weight decay (default: 1e-4)
81+
--print-freq N, -p N print frequency (default: 10)
82+
--resume PATH path to latest checkpoint (default: none)
83+
-e, --evaluate evaluate model on validation set
84+
--pretrained use pre-trained model
85+
--dali_cpu Runs CPU based version of DALI pipeline.
86+
--data_loader {pytorch,dali,dali_proxy}
87+
Select data loader: "pytorch" for native PyTorch data loader, "dali" for DALI data loader, or "dali_proxy" for PyTorch dataloader with DALI proxy preprocessing.
88+
--prof PROF Only run 10 iterations for profiling.
89+
--deterministic If enabled, random seeds are fixed to ensure deterministic results for reproducibility.
90+
--fp16-mode Enable half precision mode.
91+
--loss-scale LOSS_SCALE
92+
Loss scaling factor for mixed precision training. Default is 1.
93+
--channels-last CHANNELS_LAST
94+
Use channels-last memory format for model and data. Default is False.
95+
-t, --test Launch test mode with preset arguments

0 commit comments

Comments
 (0)