From 3ce1cbbae2f727848ad30ed651a5e998b5c36961 Mon Sep 17 00:00:00 2001 From: Mike Wasserman Date: Tue, 16 Jun 2026 10:05:28 -0700 Subject: [PATCH] Minor sampling mode updates --- README.md | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/README.md b/README.md index 27b71bb..d62ecd5 100644 --- a/README.md +++ b/README.md @@ -460,14 +460,14 @@ Note that `append()` can also cause [overflow](#tokenization-context-window-leng ### Configuration of sampling modes -For standard web page contexts, developers can specify a high-level `samplingMode` during session creation to configure the model's output variety and creativity without worrying about model-internal scalar parameters. +Developers can specify a high-level `samplingMode` during session creation to configure the model's output variety and creativity without worrying about model-internal scalar parameters. The allowed values for `samplingMode` are: -* `"most-predictable"`: For tasks requiring strict consistency and reproducibility (e.g., code generation or factual extraction). +* `"most-predictable"`: For tasks requiring strict consistency and reproducibility (e.g. testing, code generation, or content extraction). * `"predictable"`: For focused outputs with minimal variation. * `"balanced"` (default): The standard preset for most conversational interactions. -* `"creative"`: For tasks where variety and creativity are preferred over strict factual reproducibility. -* `"most-creative"`: For maximum diversity of tokens and creative brainstorming. +* `"creative"`: For tasks where variety and creativity are preferred over strict reproducibility. +* `"most-creative"`: For maximum diversity of output and creative brainstorming. Example: ```js