vshenoyTT
commented
1 month ago This tool also has a complementary web tool to visually graph operations.
GitHub Repository Deployment (available till 9/13)
Codebase
The tool is built using Streamlit for the interface, along with Pandas, NumPy, and MatPlotLib for data manipulation and graphing. The main file containing the web interface and data manipulation is streamlit_app.py. The dependencies are listed in requirements.txt, and the .streamlit file contains configurations for UI styling.
How it Works
To utilize the tool, a user selects the Grayskull or Wormhole configuration, adjusting core count for calculations accordingly. Then, the user uploads a .csv or .xlsx performance sheet. The tool automatically starts the graphing process
Results
After a few seconds, the tool displays multiple graphs for the categories of All Operations, MatMul + Conv Operations only, and Other On-Device Operations (excluding MatMul + Conv). Each category has graphs for Core Count + Utilization, Device Kernel Duration + Utilization, and Device Kernel Duration vs Utilization.
There is also a Pie Chart to show the breakdown of Operation Types in the data.
Downloaded Spreadsheet
Graphs and data are downloadable via buttons below each graph and a Download Data section at the bottom of the tool.
Run Locally
To run the tool locally, install the requirements from requirements.txt using the command pip install -r requirements.txt. Then, within the directory, run the command streamlit run streamlit_app.py. The web tool will default to opening at http://localhost:8501/, unless port 8501 is already in use.
mbahnasTT
Performance Sheet Analysis Tool
Intro
This is a web tool for easily analyzing Excel and CSV performance sheets. Primarily, it calculates FPS and Adjusted Utilization for each operation.
GitHub Repository Deployment (available till 9/13)
Codebase
The tool is built using Streamlit for the interface, with Pandas and NumPy for data manipulation and sheet formatting. The main file containing the web interface and data manipulation is
streamlit_app.py. The dependencies are listed inrequirements.txt, and the.streamlitfile contains configurations for UI styling.Usage
How it Works
To utilize the tool, a user selects the
GrayskullorWormholeconfiguration, adjusting core count for calculations accordingly. Then, the user uploads a.csvor.xlsxperformance sheet. The tool automatically starts the analysis processResults
After a few seconds, the tool displays FPS for All Operations, MatMul + Conv Operations only, and Other On-Device Operations (excluding MatMul + Conv). The user can download performance sheets filtered for any of these configurations.
Downloaded Spreadsheet
If downloaded, the modified spreadsheet appends an Adjusted Utilization column for each operation, and an FPS column for the configuration. It also shuffles the Core Count + Device Kernel Duration columns to the end for easy comparison.
Run Locally
To run the tool locally, install the requirements from requirements.txt using the command
pip install -r requirements.txt. Then, within the directory, run the commandstreamlit run streamlit_app.py. The web tool will default to opening at http://localhost:8501/, unless port 8501 is already in use.