Dataset management, annotation tools, fine-tuning frameworks, and experiment tracking.
The GitHub of AI — models, datasets, and Spaces in one platform
Essential infrastructure for the AI ecosystem. 500K+ models, 100K+ datasets, and Spaces for demos. Git-based versioning, model cards, and community features. The first place to look for any model.
ML experiment tracking, model registry, and dataset versioning
Industry standard for experiment tracking. Log metrics, visualise runs, compare experiments, and manage model lifecycle. Integrates with every major ML framework. Essential for any ML team.
2× faster LLM fine-tuning with 70% less memory
Game-changer for fine-tuning. Custom Triton kernels make QLoRA 2× faster while using 70% less VRAM. Fine-tune Llama 70B on a single 48GB GPU. Best tool for democratising fine-tuning.
Easy-to-use fine-tuning toolkit supporting YAML-based configuration
Most popular fine-tuning toolkit. YAML config makes it easy to start. Supports LoRA, QLoRA, full fine-tuning, DPO, and RLHF. Good for teams without deep ML engineering expertise.
Enterprise data labelling platform with human-in-the-loop
Leading enterprise data annotation platform. Managed labelling workforce, quality control, and RLHF services. Used by major AI labs. Best for teams needing high-volume, high-quality labels.
Open-source data labelling for text, images, audio, and video
Most flexible open-source annotation tool. Supports every modality — text, image, audio, video, HTML, and time-series. Customisable labelling interfaces. Self-hostable with ML-assisted pre-labelling.
Open-source platform for human feedback and RLHF data curation
Best open-source tool for collecting human feedback. Purpose-built for RLHF/DPO workflows. Tight Hugging Face Hub integration for publishing datasets. Essential for alignment data.
Hugging Face's parameter-efficient fine-tuning library
Essential library for parameter-efficient fine-tuning. LoRA, QLoRA, IA³, and adapters. Reduces trainable params by 99% while maintaining quality. Core dependency of every fine-tuning toolkit.
Git-based data and model version control for ML projects
Git for data. Track large datasets and models alongside code. Pipeline DAGs for reproducible experiments. Works with any storage backend. Essential for MLOps teams needing data lineage.
Lightning AI toolkit for pretraining, fine-tuning, and deploying LLMs
Clean, hackable LLM training code from Lightning AI. Supports 20+ model architectures. Good for researchers wanting to understand and modify training pipelines. Less abstraction than Axolotl.
