Make the initial_utterance customizable.

src/datasetGPT/base.py

@@ -19,7 +19,7 @@ class DatasetGenerator:
     options_configs: List[Dict[str, Any]]
     """Possible combinations of the provided options."""
     generator_index: int = 0
-    """Index of the next item be returned by the generator."""
+    """Index of the next item to be returned by the generator."""
 
     def __init__(self, config: DatasetGeneratorConfig) -> None:
         self.config = config

src/datasetGPT/cli.py

@@ -67,6 +67,12 @@ def datasetGPT() -> None:
               type=str,
               required=True,
               help="Agent role description.")
+@click.option("--initial-utterance",
+              "-u",
+              "initial_utterance",
+              type=str,
+              default="Hello!",
+              help="An utterance to be provisioned to the first agent. For many use cases a \"Hello\" is enough.")
 @click.option("--interruption",
               "-i",
               "interruption",
@@ -101,6 +107,7 @@ def conversations(
     openai_api_key: str,
     agent1: str,
     agent2: str,
+    initial_utterance: str,
     num_samples: int,
     interruption: str,
     end_phrase: str,
@@ -117,6 +124,7 @@ def conversations(
     generator_config = ConversationsGeneratorConfig(openai_api_key=openai_api_key,
                                                     agent1=agent1,
                                                     agent2=agent2,
+                                                    initial_utterance=initial_utterance,
                                                     num_samples=num_samples,
                                                     interruption=interruption,
                                                     end_phrase=end_phrase,

src/datasetGPT/conversations.py

@@ -24,6 +24,8 @@ class ConversationsGeneratorConfig:
     """Description of the first agent used to construct its system message."""
     agent2: str
     """Description of the second agent used to construct its system message."""
+    initial_utterance: str
+    """Utterance to be provisioned to the first agent."""
     num_samples: int = 1
     """Number of conversations to generate for each options combination."""
     interruption: str = "length"
@@ -95,10 +97,7 @@ def end_phrase_interruption(self, agent: str, message: str) -> None:
                 if self.config.end_phrase in message:
                     raise StopIteration()
 
-    def generate_item(
-        self,
-        initial_utterance: str = "Hello!"
-    ) -> Dict[str, Union[List[List[Any]], float, int]]:
+    def generate_item(self) -> Dict[str, Union[List[List[Any]], float, int]]:
         """Run two chains to talk with one another and record the chat history."""
         if self.generator_index >= len(self.options_configs):
             raise StopIteration()
@@ -116,7 +115,7 @@ def generate_item(
 
         utterances = []
 
-        chain1_inp = initial_utterance
+        chain1_inp = self.config.initial_utterance
         for _ in range(conversation_config["length"]):
             chain1_out = chain1.predict(input=chain1_inp)
             utterances.append(["agent1", chain1_out])