training-cost-trends

Code for the paper "The rising costs of training frontier AI models" ¹.

Execution

Using uv (recommended)

# Basic execution
uv run cost_analysis_standalone.py

# List options
uv run cost_analysis_standalone.py --help

Using pip

# Install dependencies
pip install <list of dependencies listed in pyproject.toml>

# Basic execution
python run cost_analysis_standalone.py

# List options
python run cost_analysis_standalone.py --help

Results

All results are saved in a results/ folder, with subfolders for each method and variant.

Results are currently found in results/all-methods-top_n=10-2025-03/17_exclude_finetunes_at_threshold_stage/cost_dataset_3_estimates.csv.

Data

Raw data is available in the data/ folder:

• All ML Systems - full view.csv is a snapshot of the Epoch database: https://epoch.ai/data/ai-models?view=table#explore-the-data
• Chip dataset-Grid view.csv is a snapshot of our chip database, with technical information about chips such as FLOP/s performance.
• Hardware prices.csv is a snapshot of our hardware price database, including both purchase prices and cloud rental prices.
• PCU518210518210.csv is a snapshot of this price index, used to adjust for inflation

Footnotes

1. Ben Cottier, Robi Rahman, Loredana Fattorini, Nestor Maslej, Tamay Besiroglu, and David Owen. ‘The rising costs of training frontier AI models’. ArXiv [cs.CY], 2024. arXiv. https://arxiv.org/abs/2405.21015. ↩