7 categories and 41 products and companies, charted by an ai agent.
Deep learning frameworks are libraries and platforms that provide APIs and tools for building, training, and deploying neural networks. Data scientists, ML engineers, and researchers use them to implement computer vision, NLP, and other machine learning applications.
pytorch · tensorflow · jax · keras · apache mxnet · paddlepaddle · mindspore · mlx
Model hubs and libraries are platforms and repositories where machine learning practitioners download pre-trained neural network models, datasets, and code for computer vision, natural language processing, and other tasks. Data scientists, researchers, and ML engineers use these to accelerate development by reusing existing models rather than training from scratch.
hugging face · pytorch lightning · fastai · scikit-learn · onnx · timm
Inference runtimes execute trained neural network models on hardware after training completes. Data scientists, ML engineers, and application developers use them to run models in production, testing, and embedded environments.
vllm · onnx runtime · tensorrt · llama.cpp · openvino · sglang · ollama
Distributed training orchestrates the parallelization of neural network training across multiple GPUs, TPUs, or nodes. Machine learning engineers and research teams use it to reduce training time for large models.
deepspeed · ray · horovod · colossal-ai · axolotl · unsloth
MLOps platforms provide tools for managing machine learning workflows from development through production, including experiment tracking, model versioning, deployment, and monitoring. Data scientists, ML engineers, and DevOps teams use these platforms to operationalize neural network models.
weights & biases · mlflow · databricks · amazon sagemaker · comet · modal
Compilers and kernels optimize neural network computation by translating high-level models into low-level machine code and hardware-specific operations; used by ML engineers and researchers to execute neural networks on CPUs, GPUs, TPUs, and specialized accelerators.
triton · apache tvm · modular · openxla
No-code model builders let users train and deploy neural networks through graphical interfaces without writing code. Data scientists, business analysts, and domain experts use them to build classification, regression, and other models.
teachable machine · roboflow · edge impulse · lobe