Explorer

30 technologies shown

• TruthfulQA benchmark Questions where a plausible-sounding answer is false.
• Vanishing Gradient concept Gradients shrinking toward zero as they propagate through many layers.
• llama.cpp tool CPU-first inference for quantised models, in portable C++.
• DSPy framework Builds model pipelines as programs whose prompts are optimised rather than written.
• Quantization technique Storing model weights at lower precision to cut memory and speed up inference.
• Self-RAG technique Model decides for itself when to retrieve and critiques what it retrieved.
• BGE model Embedding model family that ranks well on retrieval benchmarks.
• Chroma database Embedded vector database designed to be added to a Python application in minutes.
• Constitutional AI technique Aligning a model using written principles rather than human labels alone.
• DeepSpeed framework Training optimisation library for large models across many GPUs.
• Diffusion technique Generates data by learning to reverse a gradual noising process.
• GELU concept Activation function used in transformer feed-forward layers.
• Guardrails technique Checks around a model's input and output that constrain what it can do.
• HellaSwag benchmark Commonsense sentence completion that is trivial for people and hard for models.
• ONNX infrastructure Interchange format for models, so a trained model runs on another runtime.
• Phi model Microsoft's small models trained mainly on curated textbook-quality data.
• TPU hardware Google's purpose-built accelerator for tensor arithmetic.
• Weights & Biases tool Experiment tracking and visualisation for training runs.
• AutoGen framework Framework for multi-agent conversations where agents critique each other.
• Context Window concept Maximum number of tokens a model can attend to in one request.
• Distillation technique Training a small model to reproduce the behaviour of a larger one.
• GGUF infrastructure Quantised model format designed for fast loading and CPU inference.
• GTE model Embedding model family from Alibaba, trained for retrieval and ranking.
• JAX framework NumPy-like array library with automatic differentiation and XLA compilation.
• MLflow tool Open platform for tracking experiments and managing model lifecycle.
• MoE architecture Mixture of Experts: route each token to a few expert subnetworks instead of all parameters.
• OpenSearch database Open-source fork of Elasticsearch, with the same hybrid retrieval model.
• Red Teaming technique Adversarially probing a model for harmful or unsafe behaviour.
• SGLang tool Serving engine built around reusing shared prompt prefixes.
• Triton Inference Server tool NVIDIA's server for deploying models from several frameworks behind one API.