We investigate wether deeper or wider is better for NNs optimization under the NTK's perspective In parallel, other experiments with the same amount of content related to low rank weights NNs are available at github.com/janisaiad/MMNN.
A complete report has been done relating the theory and why all of those experiments have been done, especially getting the result page 62.
You can read it in refs/report/report.pdf
If there is an issue you can compile again the report, first run the following command (see template directory):
pdflatex -interaction=nonstopmode -output-directory=refs/wk14-report-and-transformers/stage_3a_le_bon_template/ refs/wk14-report-and-transformers/stage_3a_le_bon_template/main3.texThen, open the report in the refs/wk14-report-and-transformers/stage_3a_le_bon_template/main.pdf file.
The work is currently in progress (and will remain during a long amout of time because this topic is very rich and new) and daily/weekly commits are done.
Feel free to contact me for any information.
To install dependencies using uv, follow these steps:
-
Install uv:
macOS/Linux:
curl -LsSf https://astral.sh/uv/install.sh | shOr using wget:
wget -qO- https://astral.sh/uv/install.sh | shWindows:
powershell -ExecutionPolicy ByPass -c "irm https://astral.sh/uv/install.ps1 | iex"
Alternatively, you can install uv using:
- pipx (recommended):
pipx install uv - pip:
pip install uv - Homebrew:
brew install uv - WinGet:
winget install --id=astral-sh.uv -e - Scoop:
scoop install main/uv
- pipx (recommended):
-
Using uv in this project:
- Initialize a new virtual environment:
uv venv
- Activate the virtual environment:
source .venv/bin/activate # On Unix .venv\Scripts\activate # On Windows
- Install dependencies from requirements.txt:
uv add -r requirements.txt
- Add a new package:
uv add package_name
- Remove a package:
uv remove package_name
- Update a package:
uv pip install --upgrade package_name
- Generate requirements.txt:
uv pip freeze > requirements.txt- List installed packages:
uv pip list
If you're using macOS or Python 3, replace pip with pip3 in line 1 of launch.sh
Replace with your project folder name (which means the name of the library you are deving) in :tests/test_env.py: