A curated list of trustworthy deep learning papers. Continually updating...

🌱 EcoLogits tracks the energy consumption and environmental footprint of using generative AI models through APIs.

A curated list of awesome Green AI resources and tools to assess and reduce the environmental impacts of using and deploying AI.

Zero-dependency browser automation CLI. 70+ commands, 10 test assertions, smart commands (click/fill by text — no LLM needed). MCP server for AI agents with 500x fewer tokens. Extract, observe, script runner. 50KB, pure CDP.

[ACM e-Energy'24] Code and data for "FedZero: Leveraging Renewable Excess Energy in Federated Learning"

Use local Large Language Models (LLMs) while monitoring energy usage.

A carbon-aware ML training scheduling system that optimises machine learning workloads using real-time UK grid carbon intensity data. It simulates ML training, finds optimal low-carbon execution windows, and visualises carbon savings through an interactive multi-page Streamlit analytics dashboard.

Distributed LLM pretraining during renewable curtailment windows 🌱

A machine learning pipeline taking you from raw data to fully trained machine learning model - from data to model (d2m).

GATorch is a tool seamlessly integrated with PyTorch that enables ML developers to generate an energy consumption report. By attaching your model, the tool automatically tracks the energy consumption of your model's training and generates graphs and plots to gain in-depth insights into the energy consumption of your model.

A curated list of Green AI resources.

Know the true energy cost of your code

Research-ready and production-friendly neural network pruning for PyTorch—transparent methods, reproducible baselines, and deployment metrics to compress models for real-world use.

🌱 Estimate the environmental footprint of GenAI with EcoLogits API

A Grafana panel to monitor the carbon footprint of AWS, GCP and Azure instances. Maintains a list of data centres and their energy sources, and computes CO2e emissions by sampling energy consumption of CPU, GPU and DRAM.

Open-source energy benchmark for local LLMs. Measures Wh and CO2 using real hardware sensors (RAPL, NVML, powermetrics).

Custom implementations of backpropagation-free deep learning algorithms, rigorously benchmarked against tuned baselines with hardware-validated energy, time, memory, and CO₂e measurements.

Mapping the Landscape of AI, ML, and DL in Climate Change Research

Adaptive Sparse Training (AST): 92.1% ImageNet-100 accuracy with 61% energy savings and zero degradation. Production-ready implementations for energy-efficient deep learning with ResNet50 and modern architectures.