chore: rebase and minor changes

Signed-off-by: Dheeraj Peri <[email protected]>

Author: peri044

tools/perf/README.md

@@ -127,3 +127,12 @@ Eg:
                      --backends torch,torch_tensorrt,tensorrt \
                      --report "vgg_perf_bs1.txt"
 ```
+
+### Example models
+
+This tool benchmarks any pytorch model or torchscript module. As an example, we provide VGG16, Resnet50, EfficientNet-B0, VIT, HF-BERT models in `hub.py` that we internally test for performance.
+The torchscript modules for these models can be generated by running
+```
+python hub.py
+```
+You can refer to `benchmark.sh` on how we run/benchmark these models.

tools/perf/benchmark.sh

@@ -9,7 +9,7 @@ batch_sizes=(1 2 4 8 16 32 64 128 256)
 
 #Benchmark VGG16 model
 echo "Benchmarking VGG16 model"
-for bs in batch_sizes
+for bs in ${batch_sizes[@]}
 do
   python perf_run.py --model ${MODELS_DIR}/vgg16_scripted.jit.pt \
                      --precision fp32,fp16 --inputs="(${bs}, 3, 224, 224)" \
@@ -20,7 +20,7 @@ done
 
 # Benchmark Resnet50 model
 echo "Benchmarking Resnet50 model"
-for bs in batch_sizes
+for bs in ${batch_sizes[@]}
 do
   python perf_run.py --model ${MODELS_DIR}/resnet50_scripted.jit.pt \
                      --precision fp32,fp16 --inputs="(${bs}, 3, 224, 224)" \
@@ -31,7 +31,7 @@ done
 
 # Benchmark VIT model
 echo "Benchmarking VIT model"
-for bs in batch_sizes
+for bs in ${batch_sizes[@]}
 do
   python perf_run.py --model ${MODELS_DIR}/vit_scripted.jit.pt \
                      --precision fp32,fp16 --inputs="(${bs}, 3, 224, 224)" \
@@ -42,7 +42,7 @@ done
 
 # Benchmark EfficientNet-B0 model
 echo "Benchmarking EfficientNet-B0 model"
-for bs in batch_sizes
+for bs in ${batch_sizes[@]}
 do
   python perf_run.py --model ${MODELS_DIR}/efficientnet_b0_scripted.jit.pt \
                      --precision fp32,fp16 --inputs="(${bs}, 3, 224, 224)" \
@@ -53,7 +53,7 @@ done
 
 # Benchmark BERT model
 echo "Benchmarking Huggingface BERT base model"
-for bs in batch_sizes
+for bs in ${batch_sizes[@]}
 do
   python perf_run.py --model ${MODELS_DIR}/bert_base_uncased_traced.jit.pt \
                      --precision fp32 --inputs="(${bs}, 128)@int32;(${bs}, 128)@int32" \