Pipeline Builder now supports outputting errors in your LLM runs
You can now configure your LLM output column to contain both the output value and an error message if the row fails, offering easier debugging for LLM-related issues. With this feature, you receive both the LLM response and error message as a struct in your output column. This transparency into LLM errors makes it easier to debug processing issues and understand what is happening behind the scenes with your model. Example error messages may include:
-
“Context limit exceeded”
-
“Cannot coerce to provided output type”
-
“Input prompt is null or empty”
Example of an LLM prompt configured to include errors.
To use this feature:
-
In Pipeline Builder, open a Use LLM node and create a prompt.
-
Locate the Output type section.
-
Toggle on the Include errors checkbox.
When including LLM error messages, the node output will be a struct type, containing both the regular output value (which can be any supported output type, including a struct) and the associated error message.
Example of a struct output type (containing an array, a string, and an integer) with the included error message.
You can easily switch this feature on or off for flexible error transparency control by deselecting the Include errors box.
Learn more about using LLMs in Pipeline Builder from our documentation.