Batch Run
On this page
Purpose
Batch Run sends each row from a DataFrame to a connected language model and returns a DataFrame with one model response per row.
When to use it
- Classify a column of support tickets.
- Summarize records imported from a CSV file.
- Generate a short response for each row in a dataset.
Required setup
Connect a DataFrame and a language model. If you set Column Name, confirm that column exists in the incoming DataFrame.
How to use it
- Connect a DataFrame and a language model.
- Enter Column Name to send one column to the model, or leave it blank to send each complete row.
- Add Instructions when every row needs the same task.
- Send LLM Results to a DataFrame-aware component.
Configuration
| Setting | What it controls | Recommended starting point |
|---|---|---|
| Instructions | Adds the same instruction to every model call. | State the task and the response format. |
| Column Name | Chooses the column to process. A blank value sends every field in the row as TOML. | Start with one text column. |
| Output Column Name | Names the column that holds model responses. The default is model_response. | Keep the default unless a downstream step expects another name. |
| Enable Metadata | Adds processing metadata to output rows. The input_length field reads only text_input, so it is not reliable for another selected column. | Leave it off until you need diagnostics. |
Expected result
The output keeps the original columns and adds the response column and batch_index. A missing column produces an error. Batch Run has no component-level throttle, and most model failures stop the batch without per-row recovery. Start with a small sample before sending full rows or a large dataset to a paid model.
Reference details
Use Parser or Type Convert when the model response needs more structure. Use DataFrame Operations when the batch output needs filtering or reshaping.