bug: fix minor bugs

app.py

@@ -35,20 +35,22 @@ def get_available_tasks(parser: DatasetParser[Any]) -> list[str]:
 
 
 def format_entry_attributes(entry: ParseEntry) -> str:
-    """Format all attributes of a ParseEntry except prompt and answer."""
+    """Format all attributes of a ParseEntry except question and answer."""
     from dataclasses import fields
 
     # Get all field names from the dataclass
     field_names = [field.name for field in fields(entry)]
-    # Filter out prompt and answer
-    filtered_fields = [name for name in field_names if name not in ["prompt", "answer"]]
+    # Filter out question and answer
+    filtered_fields = [
+        name for name in field_names if name not in ["question", "answer"]
+    ]
     # Build the formatted string
     return "\n".join(f"{name}: {getattr(entry, name)}" for name in filtered_fields)
 
 
 def load_and_parse(
     parser_name: str, task_name: str | None, split_name: str | None
-) -> tuple[int, str, str, str, str, gr.Dropdown, str]:
+) -> tuple[int, str, str, str, gr.Dropdown, str]:
     """Load and parse the dataset, return the first entry and available splits."""
     try:
         parser = get_parser_instance(parser_name)
@@ -79,15 +81,14 @@ def load_and_parse(
 
         info = parser.__repr__()
         if not parsed_data:
-            return 0, "No entries found", "", "", "", split_dropdown, info
+            return 0, "", "", "", split_dropdown, info
 
         # Get the first entry
         first_entry = parsed_data[0]
 
         return (
             0,  # Return first index instead of list of indices
-            first_entry.prompt,
-            first_entry.raw_question,
+            first_entry.question,
             first_entry.answer,
             format_entry_attributes(first_entry),
             split_dropdown,
@@ -96,22 +97,22 @@ def load_and_parse(
     except Exception as e:
         # Make the error message more user-friendly and detailed
         error_msg = f"Failed to load dataset: {str(e)}\nParser: {parser_name}\nTask: {task_name}\nSplit: {split_name}"
-        return 0, error_msg, "", "", "", [], ""
+        return 0, error_msg, "", "", [], ""
 
 
 def update_entry(
     parsed_data_index: int | None, parser_name: str
-) -> tuple[str, str, str, str]:
+) -> tuple[str, str, str]:
     """Update the displayed entry based on the selected index."""
     try:
         if not parser_name:
-            return "Please select a parser first", "", "", ""
+            return "Please select a parser first", "", ""
 
         parser = get_parser_instance(parser_name)
         parsed_data = parser.get_parsed_data
 
         if not parsed_data:
-            return "No data available", "", "", ""
+            return "No data available", "", ""
 
         if parsed_data_index is None:
             # Random selection using secrets instead of random
@@ -123,13 +124,12 @@ def update_entry(
             entry = parsed_data[index]
 
         return (
-            entry.prompt,
-            entry.raw_question,
+            entry.question,
             entry.answer,
             format_entry_attributes(entry),
         )
     except Exception as e:
-        return f"Error: {str(e)}", "", "", ""
+        return f"Error: {str(e)}", "", ""
 
 
 def update_parser_options(parser_name: str) -> tuple[gr.Dropdown, gr.Dropdown, str]:
@@ -350,11 +350,8 @@ def create_interface() -> gr.Blocks:
 
                     with gr.Column(scale=2):
                         # Output displays
-                        prompt_output = gr.Textbox(
-                            label="Prompt", lines=5, show_copy_button=True
-                        )
-                        raw_question_output = gr.Textbox(
-                            label="Raw Question", lines=5, show_copy_button=True
+                        question_output = gr.Textbox(
+                            label="Question", lines=5, show_copy_button=True
                         )
                         answer_output = gr.Textbox(
                             label="Answer", lines=5, show_copy_button=True
@@ -411,8 +408,7 @@ def create_interface() -> gr.Blocks:
             inputs=[parser_dropdown, task_dropdown, split_dropdown],
             outputs=[
                 entry_index,
-                prompt_output,
-                raw_question_output,
+                question_output,
                 answer_output,
                 attributes_output,
                 split_dropdown,
@@ -430,8 +426,7 @@ def create_interface() -> gr.Blocks:
             fn=update_entry,
             inputs=[entry_index, parser_state],
             outputs=[
-                prompt_output,
-                raw_question_output,
+                question_output,
                 answer_output,
                 attributes_output,
             ],

docs/adding_new_parser.md

@@ -23,10 +23,10 @@ class YourDatasetParseEntry(HuggingFaceParseEntry):
     custom_field: str
 
     @classmethod
-    def create(cls, prompt: str, answer: str, raw_question: str,
+    def create(cls, question: str, answer: str, raw_question: str,
                raw_answer: str, task_name: str, custom_field: str) -> "YourDatasetParseEntry":
         return cls(
-            prompt=prompt,
+            question=question,
             answer=answer,
             raw_question=raw_question,
             raw_answer=raw_answer,
@@ -41,27 +41,9 @@ class YourDatasetParser(HuggingFaceDatasetParser[YourDatasetParseEntry]):
     _data_source = "huggingface/your-dataset"
     _default_task = "default"
     _task_names = ["task1", "task2", "task3"]
-    _default_system_prompt = YOUR_SYSTEM_PROMPT
 ```
 
-### 2. Define System Prompt
-
-Add your system prompt to `llmdataparser/prompts.py`:
-
-```python
-YOUR_SYSTEM_PROMPT: Final[str] = textwrap.dedent(
-    """\
-    You are an expert in [your domain]. Your task is to [describe the task].
-
-    Instructions:
-    1. [First instruction]
-    2. [Second instruction]
-    ...
-    """
-)
-```
-
-### 3. Implement Required Methods
+### 2. Implement Required Methods
 
 Your parser needs to implement these key methods:
 
@@ -78,11 +60,10 @@ def process_entry(
     raw_answer = row["answer"]
     task = task_name or self._get_current_task(row)
 
-    # Format the prompt
-    prompt = f"{self._system_prompt}\nQuestion: {raw_question}\nAnswer:"
+    question = f"Question: {raw_question}\nAnswer:"
 
     return YourDatasetParseEntry.create(
-        prompt=prompt,
+        question=question,
         answer=raw_answer,
         raw_question=raw_question,
         raw_answer=raw_answer,
@@ -115,7 +96,7 @@ def get_evaluation_metrics(self) -> list[EvaluationMetric]:
     ]
 ```
 
-### 4. Add Example Usage
+### 3. Add Example Usage
 
 Add example usage at the bottom of your parser file:
 
@@ -137,7 +118,7 @@ if __name__ == "__main__":
         print(f"Answer: {example.answer}")
 ```
 
-### 5. Create Tests
+### 4. Create Tests
 
 Create a test file `tests/test_your_dataset_parser.py`:
 
@@ -171,7 +152,6 @@ def test_process_entry():
 1. **Documentation**: Add clear docstrings and comments explaining your parser's functionality.
 1. **Error Handling**: Include appropriate error checking and validation.
 1. **Testing**: Write comprehensive tests covering different scenarios.
-1. **System Prompt**: Design your system prompt carefully to guide the model effectively.
 
 ## Examples
 
@@ -185,7 +165,6 @@ Look at existing parsers for reference:
 
 1. **Parse Entry Class**: Create a custom parse entry class if you need additional fields.
 1. **Task Names**: Define all available tasks in `_task_names`.
-1. **System Prompt**: Write clear instructions in the system prompt.
 1. **Process Entry**: Handle data extraction and formatting in `process_entry`.
 1. **Dataset Description**: Provide comprehensive dataset information.
 1. **Evaluation Metrics**: Define appropriate metrics for your dataset.

llmdataparser/prompts.py

@@ -1,6 +1,8 @@
 import textwrap
 from typing import Final
 
+# Only for reference
+
 MMLU_SYSTEM_PROMPT: Final[str] = textwrap.dedent(
     """\
     You are an expert answering multiple-choice questions. Select the single most accurate answer (A, B, C, or D) based on factual knowledge. Respond with the letter only.
@@ -44,7 +46,7 @@ IFEVAL_SYSTEM_PROMPT: Final[str] = textwrap.dedent(
 
 BBH_SYSTEM_PROMPT: Final[str] = textwrap.dedent(
     """\
-    Solve this reasoning problem and respond with only the answer (letter, True/False, or Yes/No).
+    Solve this reasoning problem step by step.
 """
 )
 

llmdataparser/tw_legal_parser.py

@@ -88,7 +88,7 @@ class TWLegalDatasetParser(HuggingFaceDatasetParser[TWLegalParseEntry]):
             ),
             citation="""
                 url={https://huggingface.co/datasets/lianghsun/tw-legal-benchmark-v1}
-            }""",
+            """,
         )
 
     def get_evaluation_metrics(self) -> list[EvaluationMetric]: