Edge Inference Benchmark

Time Series Foundation Models on Consumer GPUs: A 6GB VRAM Benchmark

A benchmark framework evaluating accuracy-efficiency tradeoffs of time series foundation models on consumer GPUs (6GB VRAM).

📦 Models Benchmarked

📊 Datasets

🚀 Quick Start

# 1. Create virtual environment
python -m venv venv
source venv/bin/activate  # Linux/Mac
# OR: venv\Scripts\activate  # Windows

# 2. Install PyTorch with CUDA first
pip install torch==2.4.1 --index-url https://download.pytorch.org/whl/cu118

# 3. Install dependencies
pip install -r requirements.txt

# 4. Download datasets
python -m src.data.download

# 5. Run benchmark
python experiments/run_baselines.py --model xgboost --dataset etth1 --runs 5
python experiments/run_foundation.py --model chronos --size small --dataset etth1

📁 Project Structure

edge_inference_benchmark/
├── configs/           # Model and dataset configurations
├── src/               # Source code
│   ├── data/          # Data loading
│   ├── models/        # Model implementations
│   ├── evaluation/    # Metrics and efficiency
│   └── visualization/ # Plotting utilities
├── experiments/       # Experiment scripts
├── results/
│   ├── figures/              # Visualization outputs
│   └── benchmark_summary.json # All metrics summary
└── requirements.txt

📈 Key Results

Accuracy vs Latency Tradeoff

Cold-Start Performance

🖥️ Hardware

Tested on:

• GPU: NVIDIA RTX 3050 Mobile (6GB VRAM)
• CUDA: 11.8
• Python: 3.12

📝 Citation

@article{author2026edge,
  title={Time Series Foundation Models on Consumer GPUs: A 6GB VRAM Benchmark},
  author={Author, Your Name},
  journal={Transactions on Machine Learning Research},
  year={2026}
}

📄 License

MIT License - see LICENSE for details.