AWS nova documentation change
Summary
Identical changes to prompting-tool-troubleshooting.md: Updated error messages, inference parameters, tool_choice examples, and removed chain-of-thought section
Security assessment
Mirror changes to the other troubleshooting guide. Modifications improve tool usage documentation but lack any security-specific context or vulnerability references. Temperature adjustment is a configuration best practice rather than security mitigation.
Diff
diff --git a/nova/latest/userguide/tools-troubleshooting.md b/nova/latest/userguide/tools-troubleshooting.md index b145cc336..876faf3b8 100644 --- a//nova/latest/userguide/tools-troubleshooting.md +++ b//nova/latest/userguide/tools-troubleshooting.md @@ -9 +9 @@ You might see different errors when working with tools and Amazon Nova models. H -`An error occurred (ModelErrorException) when calling the Converse operation: The system encountered an unexpected error during processing. Try your request again.` +`An error occurred (ModelErrorException) when calling the Converse operation: The model produced an invalid sequence as part of ToolUse. Please refer to the model tool use troubleshooting guide.` @@ -11 +11 @@ You might see different errors when working with tools and Amazon Nova models. H -**Validate your inference parameters** \- Amazon Nova models have more success calling tools when using greedy decoding. To enable greedy decoding, set the top P, top K, and temperature parameters to 1. +**Validate your inference parameters** \- Amazon Nova models have more success calling tools when using greedy decoding. To enable greedy decoding, set the temperature parameters to 0. @@ -15,2 +15 @@ You might see different errors when working with tools and Amazon Nova models. H - "topP": 1, - "temperature": 1 + "temperature": 0 @@ -45 +44 @@ If it looks like tools are being ignored by the model, make sure you are followi -You might receive a successful response from Amazon Nova, but the model does not acknowledge that the needed tool is not present. +If a tool is not being returned when expected, it is recommended to leverage the `tool_choice` API parameter. @@ -47 +46 @@ You might receive a successful response from Amazon Nova, but the model does not -**Validate your inference parameters** \- Amazon Nova models have more success calling tools when using greedy decoding. To enable greedy decoding, set the top P, top K, and temperature parameters to 1. + * **Tool** : The specified tool will be called once. @@ -48,0 +48,7 @@ You might receive a successful response from Amazon Nova, but the model does not + { + "toolChoice": { + "tool": { + "name": "name_of_tool" + } + } + } @@ -50,3 +56,6 @@ You might receive a successful response from Amazon Nova, but the model does not - inferenceConfig={ - "topP": 1, - "temperature": 1 + * **Any** : One of the provided tools will be called at least once. + + { + "toolChoice": { + "any": {} + } @@ -55 +63,0 @@ You might receive a successful response from Amazon Nova, but the model does not - additionalModelRequestFields={"inferenceConfig": {"topK": 1}} @@ -57 +65 @@ You might receive a successful response from Amazon Nova, but the model does not -**Review the system prompt** \- To improve the accuracy of tool calling, Amazon Nova uses chain-of-thought reasoning when calling a tool. You will see this output in the response in <thinking> tags. We do not recommended trying to remove this functionality. Instead, we recommend that you drop the output if you do not need it in your application. +