提交 96a56136 编写于 作者: M mir-of

add readme

上级 72c4858f
# OneFlow-Benchmark
OneFlow models for benchmarking.
## Training
### cnns
## CNNs
### Train
* 1 node, 1 gpu:
```
python3 cnn_benchmark/of_cnn_train_val.py \
......@@ -40,10 +40,24 @@ OneFlow models for benchmarking.
--use_fp16=False \
--use_boxing_v2=True \
--model="resnet50"
```
### Validation
```
python3 cnn_benchmark/of_cnn_val.py \
--model_load_dir=output/snapshots_0323 \
--val_data_dir=$DATA_ROOT/validation \
--val_data_part_num=256 \
--gpu_num_per_node=4 \
--loss_print_every_n_iter=20 \
--val_batch_size_per_device=125 \
--model="resnet50"
```
### bert pretrain
## BERT
### Pretrain
* 1 node, 1gpu:
* bert base:
```
python3 bert_benchmark/run_pretraining.py \
......
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册