Deep Learning Systems

Time-Series Demand Forecasting for LLM Inference Serving

This repository contains a notebook-based analysis of multi-step demand forecasting for LLM inference workloads using the Azure LLM Inference Trace 2024 dataset. The project compares LSTM and Transformer models against simpler statistical baselines and includes saved figures, trained model artifacts, and a written report.

GitHub repository: https://github.com/sam-andaluri/deep-learning-project

1. Cloning repo

This repo stores compressed dataset files (.csv.gz) with Git LFS.

Install git-lfs cli using https://git-lfs.com or on MacOS brew install git-lfs

Run the following commands to get the datafiles.

git lfs install
git clone https://github.com/sam-andaluri/deep-learning-project.git
cd deep-learning-project
git lfs pull
gunzip -k data/azure_llm_2024/*.gz

2. Install uv

Install uv with the standalone installer:

curl -LsSf https://astral.sh/uv/install.sh | sh

Or install it with Homebrew on macOS:

brew install uv

Confirm the installation:

uv --version

3. Prerequisites

Make sure the following tools are available on your system:

• Python 3.8 or higher
• Jupyter with nbconvert
• Pandoc for PDF generation

Quick checks:

python --version
python -m jupyter nbconvert --version
pandoc --version

4. Create and Activate a Virtual Environment

From the project folder:

cd deep-learning-project
uv venv .venv
source .venv/bin/activate

5. Install Dependencies

Install the pinned dependencies from requirements.txt:

uv pip install -r requirements.txt

6. Run the Notebook

Open and run deep_learning.ipynb in Jupyter:

jupyter notebook deep_learning.ipynb

Or using JupyterLab:

jupyter lab deep_learning.ipynb

Run all cells from top to bottom (Cell > Run All).

7. Generate the Report PDF

If you update the report markdown, regenerate the PDF with:

pandoc Deep_Learning_Systems_Analysis_Report.md -o Deep_Learning_Systems_Analysis_Report.pdf

8. Generate Final requirements.txt

After running the notebook, generate the exact package versions:

pip freeze > requirements.txt

References

Astral. (n.d.). Installing uv. uv documentation. Retrieved April 14, 2026, from https://docs.astral.sh/uv/getting-started/installation/

Astral. (n.d.). Pip interface. uv documentation. Retrieved April 14, 2026, from https://docs.astral.sh/uv/pip/

Astral. (n.d.). Using environments. uv documentation. Retrieved April 14, 2026, from https://docs.astral.sh/uv/pip/environments/

Microsoft Azure. (n.d.). Azure LLM Inference Trace 2024 [Data set]. Retrieved April 14, 2026, from https://github.com/Azure/AzurePublicDataset

Stojkovic, U., et al. (2025). DynamoLLM: Designing LLM Inference Clusters for Performance and Energy Efficiency. IEEE International Symposium on High-Performance Computer Architecture (HPCA '25).

Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural Computation, 9(8), 1735-1780.

Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., & Polosukhin, I. (2017). Attention is all you need. Advances in Neural Information Processing Systems, 30.