SIGNAL//DESK
AI/MLsrc: Roost FA glossary v2

Context window

How much text an LLM can see at once — including your prompt and its own reply. If your client chat history exceeds this limit, the model will lose track of earlier context, potentially leading to a loss of decision survivability.

The total capacity of tokens a model can process in a single interaction. This includes the system prompt, the conversation history, and the current input. Once exceeded, the model "forgets" the earliest parts of the conversation.

The total capacity of tokens a model can process in a single interaction. This includes the system prompt, the conversation history, and the current input. Once exceeded, the model "forgets" the earliest parts of the conversation.

evolution

  1. 2017-06 · history
    Transformer Architecture

    The 'Attention Is All You Need' paper introduced the self-attention mechanism, establishing the fixed-length token limit inherent to Transformer models.

  2. 2020-05 · history
    GPT-3 Release

    OpenAI launched GPT-3 with a 2,048-token context window, setting the standard for large-scale generative language models of that era.

  3. 2023-03 · history
    GPT-4 Expansion

    OpenAI introduced GPT-4 with an 8k context window, later expanded to 32k, significantly increasing the amount of information models could process simultaneously.

  4. 2023-06 · history
    Context Window Scaling

    Anthropic released Claude with a 100k token context window, shifting the industry focus toward massive document analysis capabilities.

  5. 2024-02 · history
    Gemini 1.5 Pro

    Google introduced a 1-million-token context window, enabling the processing of vast datasets, entire codebases, or long-form video in a single prompt.

seen in events


← all terms