Contents x
    4. Adjust Context Size
      • PDF
      Contents

      4. Adjust Context Size

        • PDF
        Article summary

        • How do I train my AI with Context Size?

        • Here you can find out what it means when you turn this value up or down. Feel free to play around with the context size and see what happens.

        🎒 What is the Context Size?

        • Think of "Context Size" as the backpack of AI

        • When you ask your AI a question, it packs text passages from your product data / product texts into its backpack to build the best possible answer.

          1. The larger the "Context Size" value, the more information she can take with her.

          2. The smaller the value, the less information it has with it – it has less information available for an answer, but it is faster.

        Standard configuration (recommended)

        1. Set the Context Size to 20 .

        2. Note: Navigate to Algorithm in the sidebar on the left. In the "Number of Recommendations Shown" field, set the same number as in the Context Size (also 20)

        3. For your chat functionality, set the number of products you have in the system in the "Number of Recommendations Returned" field.

        Example from everyday life

        You ask: What is the difference between laptop A and laptop B?

        Now the AI looks in your product data for suitable places that have something to do with both laptops – e.g. technical data, descriptions, ratings.

        If you set the Context Size small (e.g. 5):

        • The AI only gets a small part of the analysis

        • Maybe the first 5 relevant data sets only contain something about laptop A – but nothing more about laptop B.

        • The answer seems one-sided, incomplete.

        • BUT: Extreme speed in small data analyses

        If you set the Context Size larger (e.g. 100):

        Now it can analyze more sections at once:

        • Comparison tables

        • Battery life of laptop A

        • Weight of laptop B

        • Price-performance conclusion

        💡 Note on Context Size

        • Note that in the settings, the "Show Recommendation" should be equal to the Context Size. The maximum context size must not be greater than the number of products displayed in "Recommendation show".

        • If speed is lacking, you can choose a different generative model that fits your company's data protection requirements. Note that in this case, you should reset the feedback you have given so far and start over.

        🎯 Rule of thumb

        • The more background knowledge the AI needs, the larger the backpack (context size) should be.

        ⚠️ Attention

        • A backpack that is too large can also bring ballast, as the quality of the answer could suffer if there are too many texts.

        • The answer can seem more balanced, well-founded and detailed, but with large amounts of data, the AI also works more slowly.

        🛠 When does a large context size help you?

        • You want to make a comparison = Higher Context Size

        • The information is spread over many data chunks = Higher Context Size

        • You need a well-founded answer with details = Higher Context Size

        • You ask a very specific question about a data chunk = smaller context size is enough

        Was this article helpful?
        What's Next