LayersLayer 4 of 15
Training and fine-tuning
Training and fine-tuning covers the frameworks, libraries and platforms used to train a model from scratch or adapt an existing one to a task, and the tools that track those experiments.
The categories
12 categories and 315 products in Training and fine-tuning.
CategoryWhat it coversProducts
AutoML & hyperparameter optimizationAutoML & hyperparameter optimization covers automatic model, architecture and hyperparameter search, including full AutoML systems (Optuna, AutoGluon, FLAML, H2O AutoML).25
Computer vision toolkits & platformsComputer vision toolkits & platforms covers libraries and end-to-end platforms for building, training and deploying custom detection, segmentation and tracking models to cloud or edge.38
Deep learning & ML frameworksDeep learning & ML frameworks covers core libraries to define, train and run models: tensor/autograd frameworks and classical or gradient-boosting ML libraries.40
Distributed training frameworksDistributed training frameworks covers frameworks for training one model across many GPUs or many machines.24
Experiment trackingExperiment tracking covers tools that record training runs, their results and their output files.18
Feature storesFeature stores covers systems that manage and serve the input data models use, for both training and live predictions.15
Federated & privacy-preserving learningFederated & privacy-preserving learning covers frameworks for training across data silos or devices without centralizing data: federated learning, split learning, secure aggregation.22
Fine-tuning platformsFine-tuning platforms covers managed services that adapt a base model to your data.21
ML workflow / MLOps platformsML workflow / MLOps platforms covers end-to-end pipelines, orchestration and lifecycle management for ML.32
Open-source fine-tuning librariesOpen-source fine-tuning libraries covers self-run libraries for fine-tuning models on your own GPUs (LoRA/QLoRA, SFT, recipe toolkits); the code counterpart to managed fine-tuning platforms.30
RL / post-training librariesRL / post-training libraries covers RLHF, DPO and RL trainer libraries (verl, OpenRLHF, TRL's RL trainers).20
RL environments & gymsRL environments & gyms covers built, sold or hosted interactive environments, tasks and verifiers used to train (and test) agents and models with reinforcement learning.30
Questions
Training and fine-tuning, answered.
What is the Training and fine-tuning layer of AI infrastructure?Training and fine-tuning covers the frameworks, libraries and platforms used to train a model from scratch or adapt an existing one to a task, and the tools that track those experiments.
What does the Training and fine-tuning layer include?It has 12 categories: AutoML & hyperparameter optimization, Computer vision toolkits & platforms, Deep learning & ML frameworks, Distributed training frameworks, Experiment tracking, Feature stores, Federated & privacy-preserving learning, Fine-tuning platforms, ML workflow / MLOps platforms, Open-source fine-tuning libraries, RL / post-training libraries and RL environments & gyms.
How many Training and fine-tuning products does Unfragile track?315 products across 12 categories, each category with its own page.
What works here
The three products on this layer.
SearchLive
RecommendNot live yet.
RouteNot live yet.