Make scripts executable. Be consistent about execution.

README.md

@@ -2,7 +2,8 @@
 We supply a small microbenchmarking script for PyTorch training on ROCm.
 
 To execute:
-`python micro_benchmarking_pytorch.py --network <network name> [--batch-size <batch size> ] [--iterations <number of iterations>] [--fp16 <0 or 1> ] [--dataparallel|--distributed_dataparallel] [--device_ids <comma separated list (no spaces) of GPU indices (0-indexed) to run dataparallel/distributed_dataparallel api on>]`
+
+`./micro_benchmarking_pytorch.py --network <network name> [--batch-size <batch size> ] [--iterations <number of iterations>] [--fp16 <0 or 1> ] [--dataparallel|--distributed_dataparallel] [--device_ids <comma separated list (no spaces) of GPU indices (0-indexed) to run dataparallel/distributed_dataparallel api on>]`
 
 Possible network names are: `alexnet`, `densenet121`, `inception_v3`, `resnet50`, `resnet101`, `SqueezeNet`, `vgg16` etc.
 
@@ -14,17 +15,18 @@ For mGPU runs, `--distributed_dataparallel` with 1 GPU per process is recommende
 Eg. 
 for a 1-GPU resnet50 run:
 ```
-python3 micro_benchmarking_pytorch.py --network resnet50
+./micro_benchmarking_pytorch.py --network resnet50
 ```
 for a 2-GPU run on a single node:
 ```
-python3 micro_benchmarking_pytorch.py --device_ids=0 --network resnet50 --distributed_dataparallel --rank 0 --world-size 2 --dist-backend nccl --dist-url tcp://127.0.0.1:4332 &
-python3 micro_benchmarking_pytorch.py --device_ids=1 --network resnet50 --distributed_dataparallel --rank 1 --world-size 2 --dist-backend nccl --dist-url tcp://127.0.0.1:4332 &
+./micro_benchmarking_pytorch.py --device_ids=0 --network resnet50 --distributed_dataparallel --rank 0 --world-size 2 --dist-backend nccl --dist-url tcp://127.0.0.1:4332 &
+./micro_benchmarking_pytorch.py --device_ids=1 --network resnet50 --distributed_dataparallel --rank 1 --world-size 2 --dist-backend nccl --dist-url tcp://127.0.0.1:4332 &
 ```
 Specify any available port in the `dist-url`.
 
 To run FlopsProfiler (with deepspeed.profiling.flops_profiler imported):
-`python micro_benchmarking_pytorch.py --network resnet50 --amp-opt-level=2 --batch-size=256 --iterations=20 --flops-prof-step 10`
+
+`./micro_benchmarking_pytorch.py --network resnet50 --amp-opt-level=2 --batch-size=256 --iterations=20 --flops-prof-step 10`
 
 ## Performance tuning
 If performance on a specific card and/or model is found to be lacking, typically some gains can be made by tuning MIOpen. For this, `export MIOPEN_FIND_ENFORCE=3` prior to running the model. This will take some time if untuned configurations are encountered and write to a local performance database. More information on this can be found in the [MIOpen documentation](https://rocmsoftwareplatform.github.io/MIOpen/doc/html/perfdatabase.html).
@@ -53,14 +55,15 @@ Added the `--compile` option opens up PyTorch 2.0 capabilities, which comes with
 With the required `--compile` option, these additional options are now available from the command line with the `--compileContext` flag. Here are a few examples:
 
 ```bash
-python micro_benchmarking_pytorch.py --network resnet50 --compile # default run
+./micro_benchmarking_pytorch.py --network resnet50 --compile # default run
 ```
 
 ```bash
-python micro_benchmarking_pytorch.py --network resnet50 --compile --compileContext "{'mode': 'max-autotune', 'fullgraph': 'True'}"
+./micro_benchmarking_pytorch.py --network resnet50 --compile --compileContext "{'mode': 'max-autotune', 'fullgraph': 'True'}"
 ```
 
 ```bash
-python micro_benchmarking_pytorch.py --network resnet50 --compile --compileContext "{'options': {'static-memory': 'True', 'matmul-padding': 'True'}}"
+./micro_benchmarking_pytorch.py --network resnet50 --compile --compileContext "{'options': {'static-memory': 'True', 'matmul-padding': 'True'}}"
 ```
+
 Note: you cannot pass the `mode` and `options` options together.

TorchTensorOpsBench/README.md

@@ -1,11 +1,14 @@
 To run the microbenchmark for an op:
+
 ```
-python torch_tensor_ops_bench.py --op <op_name>
+./torch_tensor_ops_bench.py --op <op_name>
 ```
 
 The script also takes optional arguments:
+
 ```
 --dtype [=fp32 | fp16 | bf16]
 --device [=cuda | cpu]
 --input-dim dims separated by '-', default "64-1024-1024"
 --op-type [=None | binary(for a binary op)]
+```

TorchTensorOpsBench/run.sh

@@ -1,18 +1,15 @@
-#!/usr/bin/env bash
+#!/bin/bash
 
 # run model ops in fp32, fp16 and bf16
-printf "########## Running model ops with fp32 type ###########\n"
-python3 torch_tensor_ops_bench.py --run-model-ops --dtype fp32 |& tee model_ops_fp32.log
-printf "\n########## Running model ops with fp16 type ###########\n"
-python3 torch_tensor_ops_bench.py --run-model-ops --dtype fp16 |& tee model_ops_fp16.log
-printf "\n########## Running model ops with bf16 type ###########\n"
-python3 torch_tensor_ops_bench.py --run-model-ops --dtype bf16 |& tee model_ops_bf16.log
+echo -e "########## Running model ops with fp32 type ###########"
+./torch_tensor_ops_bench.py --run-model-ops --dtype fp32 |& tee model_ops_fp32.log
+echo -e "\n########## Running model ops with fp16 type ###########"
+./torch_tensor_ops_bench.py --run-model-ops --dtype fp16 |& tee model_ops_fp16.log
+echo -e "\n########## Running model ops with bf16 type ###########"
+./torch_tensor_ops_bench.py --run-model-ops --dtype bf16 |& tee model_ops_bf16.log
 
 # run predefined ops with generic tensor size of 64-1024-1024
-printf "\n########## Running pre-defined ops with fp32 type ###########\n"
-python3 torch_tensor_ops_bench.py --run-predefined --dtype fp32 |& tee predefined_ops_fp32.log
-printf "\n########## Running pre-defined ops with fp16 type ###########\n"
-python3 torch_tensor_ops_bench.py --run-predefined --dtype fp16 |& tee predefined_ops_fp16.log
-
-printf "Done\n"
-
+echo -e "\n########## Running pre-defined ops with fp32 type ###########"
+./torch_tensor_ops_bench.py --run-predefined --dtype fp32 |& tee predefined_ops_fp32.log
+echo -e "\n########## Running pre-defined ops with fp16 type ###########"
+./torch_tensor_ops_bench.py --run-predefined --dtype fp16 |& tee predefined_ops_fp16.log

TorchTensorOpsBench/torch_tensor_ops_bench.py

@@ -1,3 +1,5 @@
+#!/usr/bin/env python3
+
 import torch
 import torch.nn as nn
 

micro_benchmarking_pytorch.py

@@ -1,3 +1,5 @@
+#!/usr/bin/env python3
+
 import torch
 import torchvision
 import random

shufflenet.py

@@ -1,3 +1,5 @@
+#!/usr/bin/env python3
+
 import os
 import sys
 import torch

shufflenet_v2.py

@@ -1,3 +1,5 @@
+#!/usr/bin/env python3
+
 import os
 import sys
 import torch

xception.py

@@ -1,3 +1,5 @@
+#!/usr/bin/env python3
+
 import math
 import torch.nn as nn
 import torch.nn.functional as F