prompted-generator

name	prompted-generator
description	Reference documentation for the PromptedGenerator operator. Covers constructor parameters, run() signature, actual row-processing behavior, and pipeline usage notes. Use when: integrating PromptedGenerator into a DataFlow pipeline for single-field LLM generation.
trigger_keywords	["PromptedGenerator","prompted-generator","single-field generation","LLM generation"]
version	1.0.0

PromptedGenerator Operator Reference

PromptedGenerator is DataFlow's basic single-field LLM generation operator.

For each row in the current DataFrame, it reads the value from input_key. If that value is truthy, it builds one LLM input as:

user_prompt + str(row[input_key])

It then calls llm_serving.generate_from_input(...) in batch and writes the generated results into output_key.

1. Import

from dataflow.operators.core_text import PromptedGenerator

2. Constructor

PromptedGenerator(
    llm_serving,                               # required
    system_prompt="You are a helpful agent.",  # optional
    user_prompt="",                            # optional
    json_schema=None,                          # optional
)

Parameter	Required	Default	Description
llm_serving	Yes	None	LLM service object implementing generate_from_input(...)
system_prompt	No	"You are a helpful agent."	System prompt passed to the serving layer
user_prompt	No	""	Prefix prepended directly before each valid row's input text
json_schema	No	None	Optional schema forwarded to the serving layer

3. run() Signature

op.run(
    storage=self.storage.step(),
    input_key="raw_content",
    output_key="generated_content",
)
# returns: output_key

Parameter	Required	Default	Description
storage	Yes	None	Current operator-step storage object
input_key	No	"raw_content"	Column read from the current DataFrame
output_key	No	"generated_content"	Column written back to the DataFrame

4. Actual Execution Logic

The operator performs the following steps:

1. Read the current DataFrame from storage.
2. Iterate over the DataFrame row by row.
3. For each row, read raw_content = row.get(input_key, "").
4. Only if raw_content is truthy, append user_prompt + str(raw_content) to the batch LLM input list.
5. Call:

llm_serving.generate_from_input(
    user_inputs=llm_inputs,
    system_prompt=system_prompt,
    json_schema=json_schema,
)

6. Write the generated list into dataframe[output_key].
7. Persist the updated DataFrame through storage.write(dataframe).
8. Return output_key.

5. Important Rules

1. input_key must exist in the current DataFrame.
2. user_prompt is a plain string prefix, not a template engine.
3. The implementation skips rows whose input_key value is empty, None, or otherwise falsy when building LLM inputs.

6. Typical Usage

from dataflow.operators.core_text import PromptedGenerator

prompted_gen = PromptedGenerator(
    llm_serving=self.llm_serving,
    system_prompt="You are a helpful agent.",
    user_prompt="Summarize the following content in one sentence:\n",
)

prompted_gen.run(
    storage=self.storage.step(),
    input_key="raw_content",
    output_key="summary",
)

7. Return Value

return output_key

This is intended for downstream operator chaining.