Update README

Author: pgworld

Fine_tuning/PipeStore/main.py

@@ -54,7 +54,6 @@
 ===================================================
                     ASPLOS 2024
 ---------------------------------------------------'''
-print(start_message)
 
 client = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
 

Fine_tuning/Tuner/server.py

@@ -28,12 +28,11 @@
 ===================================================
                     ASPLOS 2024
 ---------------------------------------------------'''
-print(start_message)
 
 overall_start = time.perf_counter()
 
 ### Set communication to clients(SOFA SSDs)
-comm = CommUnit(args.split_number, args.num_of_client, args.port, client=CLIENT)
+comm = CommUnit(args.num_of_run, args.num_of_client, args.port, client=CLIENT)
 comm.get_SSD_path()
 comm.send_message(f'dir:{os.path.abspath(os.path.join(os.path.realpath(__file__), os.pardir))}', True)
 comm.get_message("Feature Extraction Started")

Offline_inference/resnet50.onnx

@@ -153,6 +153,15 @@ trtexec --onnx=resnet50.onnx --workspace=8192 --saveEngine=resnet50.engine --bui
 ```
 - The script uses a command-line argument for the port if provided; otherwise, it defaults to 25258.
 
+3. Review the inference results. In this test, we provide two pieces of information: the elapsed time for the feature extraction time/throughput (executed in PipeStore) and the overall fine-tuning time.
+
+```
+# In AWS setting
+Feature extraction time (sec):               31.360074814000654
+Feature extraction throughput (image/sec):   1913.2607417509444
+Overall fine-tuning time (sec):              75.19124622900017
+```
+
 ## Installation & Execution (Offline Inference)
 
 For offline inference evaluation, we provide a simple test code. This code can be executed on the PipeStore side.
@@ -161,7 +170,7 @@ For offline inference evaluation, we provide a simple test code. This code can b
 
 The prerequisites for offline inference are almost identical to those required for fine-tuning.
 
-1. Follow steps 1-7 of the fine-tuning guide.
+1. Follow steps 1-7 of the fine-tuning guide (Only PipeStore setup is needed).
 
 2. Install the deflate module by following the instructions below:
 ```
@@ -177,6 +186,12 @@ The prerequisites for offline inference are almost identical to those required f
 ../Offline_inference# unzip inference_dataset.zip 
 ```
 
+4. Compile the model specifically for your GPU
+```
+trtexec --onnx=resnet50.onnx --workspace=8192 --saveEngine=resnet50.engine --buildOnly --inputIOFormats=fp16:chw --outputIOFormats=fp16:chw --fp16
+```
+
+
 ### Running the System (Takes < 1 min to run)
 
 1. Execute the following command: the first argument specifies the path to the model's engine, and the second argument specifies the dataset path.