Skip to content

PRIOR_ART.md

Latest commit

 

History

History
60 lines (39 loc) · 3.13 KB

PRIOR_ART.md

File metadata and controls

60 lines (39 loc) · 3.13 KB

Prior Art

Commit Carbon builds on and differentiates from the following prior work.

Academic Research

Strubell, E., Ganesh, A., McCallum, A. (2019). "Energy and Policy Considerations for Deep Learning in NLP." ACL 2019. Covers training emissions for NLP models. Established early awareness of AI energy costs. How we use: context for training emissions (out of scope for Commit Carbon but referenced).

Patterson, D., et al. (2021). "Carbon Emissions and Large Neural Network Training." arXiv preprint. Carbon accounting methodology for large model training. How we use: methodological reference for accounting approaches.

Luccioni, A. S., et al. (2023). "Power Hungry Processing: Watts Driving the Cost of AI Deployment." Inference emissions for deployed models. Critical baseline for our per-token energy estimates. How we use: primary source for inference energy factors.

Li, P., et al. (2023). "Making AI Less Thirsty: Uncovering and Addressing the Secret Water Footprint of AI Models." Water consumption from AI inference cooling. How we use: noted for future v2.0 water footprint extension.

Existing Tools

Code Carbon (github.com/mlco2/codecarbon) Measures training and inference emissions for ML models. Does not cover AI coding tool usage at commit level. Different scope entirely. Commit Carbon is complementary.

ML CO2 Impact Calculator (mlco2.github.io/impact) Calculates emissions from model training runs. Scope: model training only, not coding tool invocation.

Green Software Foundation Carbon Aware SDK Enables software to schedule workloads based on grid carbon intensity. Data aware computing tool, not a measurement framework for existing emissions. We reference their work on carbon aware scheduling.

Electricity Maps (electricitymaps.com) Real-time grid carbon intensity data by zone. Data source that Commit Carbon integrates as an optional grid adapter. Not a measurement framework.

WattTime (watttime.org) Real-time marginal emissions data. Data source that Commit Carbon integrates as an optional grid adapter.

Standards and Frameworks

GHG Protocol (ghgprotocol.org) Defines Scope 1/2/3 emissions framework. Commit Carbon aligns with Scope 3 Category 1 (Purchased Goods and Services). We do not copy GHG Protocol text.

CDP (cdp.net) Disclosure framework for environmental impact. Commit Carbon generates CDP compatible outputs. We do not copy CDP questionnaire text.

SBTi (sciencebasedtargets.org) Science Based Targets initiative for emissions reduction. Commit Carbon data can inform SBTi target setting but does not implement SBTi methodology.

ISO 14064 International standard for greenhouse gas accounting. Commit Carbon methodology is designed to be ISO 14064 compatible.

What Makes Commit Carbon Different

None of the above tools or standards provide:

  1. Commit level granularity for AI coding emissions
  2. CSRD compatible disclosure format specifically for AI coding
  3. Aggregation from ai-attestation data
  4. Regional grid integration for coding tool emissions
  5. Methodology audit and third party verification support

Commit Carbon is the first tool to occupy this specific intersection.