Обзор

30 technologies shown

• Transformer architecture Attention-based neural architecture that replaced recurrence with self-attention.
• RAG technique Retrieval-Augmented Generation: ground a model answer in documents fetched at query time.
• LLM concept Large language model: a Transformer trained on text at a scale that yields general capability.
• GPT model Family of decoder-only Transformer language models trained with a generative objective.
• Qdrant database Open-source vector database written in Rust, with payload filtering.
• LangChain framework Framework for composing language models with retrieval, tools and memory.
• Embedding concept Dense numeric vector representing the meaning of a piece of text.
• Attention concept Mechanism that weights each part of an input by its relevance to every other part.
• GPU hardware A processor built for massive parallelism, which is what neural network maths needs.
• Llama model Open-weight language model family from Meta, released for local and self-hosted use.
• LlamaIndex framework Data framework focused on ingesting, indexing and querying private knowledge.
• A100 hardware NVIDIA's datacentre GPU that made large-model training routine.
• Hugging Face Transformers library Reference implementations and pretrained weights for transformer models, behind one API.
• Pinecone database Fully managed proprietary vector database with no self-hosted option.
• PyTorch framework The dominant deep learning framework, built around dynamic computation graphs.
• NVIDIA CUDA platform NVIDIA's parallel computing platform, the layer most AI software is written against.
• H100 hardware NVIDIA's datacentre GPU that became the default for training large models.
• Common Crawl dataset Non-profit crawl of the public web, and the base of most training corpora.
• Docker infrastructure Packages an application and its dependencies into a portable image.
• Milvus database Open-source distributed vector database designed for billion-scale collections.
• BERT model Encoder-only Transformer pre-trained with masked language modelling.
• Kubernetes infrastructure Orchestrates containers across machines, handling scheduling and recovery.
• Qwen model Alibaba's open-weight model family, strong on multilingual and coding tasks.
• Weaviate database Open-source vector database with hybrid keyword and vector search built in.
• vLLM tool High-throughput inference server built around paged attention and continuous batching.
• Backpropagation technique Computing gradients through a network by applying the chain rule backwards.
• Code Generation application Producing or completing source code from a description or surrounding context.
• MMLU benchmark Multiple-choice benchmark across 57 academic subjects.
• Mistral model European open-weight model family known for efficiency at small parameter counts.
• AI Agent application A system that plans and takes actions with tools, rather than only answering.