chore(wren-ai-service): minor updates

wren-ai-service/src/pipelines/generation/followup_sql_generation.py

@@ -13,8 +13,10 @@
 from src.pipelines.generation.utils.sql import (
     SQL_GENERATION_MODEL_KWARGS,
     SQLGenPostProcessor,
+    calculated_field_instructions,
     construct_ask_history_messages,
     construct_instructions,
+    metric_instructions,
     sql_generation_system_prompt,
 )
 from src.pipelines.retrieval.sql_functions import SqlFunction
@@ -34,9 +36,12 @@
     {{ document }}
 {% endfor %}
 
-{% if instructions %}
-### INSTRUCTIONS ###
-{{ instructions }}
+{% if calculated_field_instructions %}
+{{ calculated_field_instructions }}
+{% endif %}
+
+{% if metric_instructions %}
+{{ metric_instructions }}
 {% endif %}
 
 {% if sql_functions %}
@@ -56,9 +61,15 @@
 {% endfor %}
 {% endif %}
 
+{% if instructions %}
+### USER INSTRUCTIONS ###
+{% for instruction in instructions %}
+{{ loop.index }}. {{ instruction }}
+{% endfor %}
+{% endif %}
+
 ### QUESTION ###
 User's Follow-up Question: {{ query }}
-Current Time: {{ current_time }}
 
 ### REASONING PLAN ###
 {{ sql_generation_reasoning }}
@@ -87,11 +98,12 @@ def prompt(
         sql_generation_reasoning=sql_generation_reasoning,
         instructions=construct_instructions(
             configuration,
-            has_calculated_field,
-            has_metric,
             instructions,
         ),
-        current_time=configuration.show_current_time(),
+        calculated_field_instructions=calculated_field_instructions
+        if has_calculated_field
+        else "",
+        metric_instructions=metric_instructions if has_metric else "",
         sql_samples=sql_samples,
         sql_functions=sql_functions,
     )

wren-ai-service/src/pipelines/generation/followup_sql_generation_reasoning.py

@@ -12,34 +12,14 @@
 from src.core.provider import LLMProvider
 from src.pipelines.generation.utils.sql import (
     construct_instructions,
+    sql_generation_reasoning_system_prompt,
 )
 from src.web.v1.services import Configuration
 from src.web.v1.services.ask import AskHistory
 
 logger = logging.getLogger("wren-ai-service")
 
 
-sql_generation_reasoning_system_prompt = """
-### TASK ###
-You are a helpful data analyst who is great at thinking deeply and reasoning about the user's question and the database schema, and you provide a step-by-step reasoning plan in order to answer the user's question.
-
-### INSTRUCTIONS ###
-1. Think deeply and reason about the user's question and the database schema, and should consider the user's query history.
-2. Give a step by step reasoning plan in order to answer user's question.
-3. The reasoning plan should be in the language same as the language user provided in the input.
-4. Make sure to consider the current time provided in the input if the user's question is related to the date/time.
-5. Don't include SQL in the reasoning plan.
-6. Each step in the reasoning plan must start with a number, a title(in bold format in markdown), and a reasoning for the step.
-7. If SQL SAMPLES are provided, make sure to consider them in the reasoning plan.
-8. If INSTRUCTIONS section is provided, please follow them strictly.
-9. Do not include ```markdown or ``` in the answer.
-10. A table name in the reasoning plan must be in this format: `table: <table_name>`.
-11. A column name in the reasoning plan must be in this format: `column: <table_name>.<column_name>`.
-
-### FINAL ANSWER FORMAT ###
-The final answer must be a reasoning plan in plain Markdown string format
-"""
-
 sql_generation_reasoning_user_prompt_template = """
 ### DATABASE SCHEMA ###
 {% for document in documents %}
@@ -57,8 +37,10 @@
 {% endif %}
 
 {% if instructions %}
-### INSTRUCTIONS ###
-{{ instructions }}
+### USER INSTRUCTIONS ###
+{% for instruction in instructions %}
+{{ loop.index }}. {{ instruction }}
+{% endfor %}
 {% endif %}
 
 ### User's QUERY HISTORY ###
@@ -71,7 +53,6 @@
 
 ### QUESTION ###
 User's Question: {{ query }}
-Current Time: {{ current_time }}
 Language: {{ language }}
 
 Let's think step by step.
@@ -98,7 +79,6 @@ def prompt(
             configuration=configuration,
             instructions=instructions,
         ),
-        current_time=configuration.show_current_time(),
         language=configuration.language,
     )
 

wren-ai-service/src/pipelines/generation/intent_classification.py

@@ -27,11 +27,12 @@
 
 ### Instructions ###
 - **Follow the user's previous questions:** If there are previous questions, try to understand the user's current question as following the previous questions.
-- **Consider Both Inputs:** Combine the user's current question and their previous questions together to identify the user's true intent.
+- **Consider Context of Inputs:** Combine the user's current question, their previous questions, and the user's instructions together to identify the user's true intent.
 - **Rephrase Question":** Rewrite follow-up questions into full standalone questions using prior conversation context."
 - **Concise Reasoning:** The reasoning must be clear, concise, and limited to 20 words.
 - **Language Consistency:** Use the same language as specified in the user's output language for the rephrased question and reasoning.
 - **Vague Queries:** If the question is vague or does not related to a table or property from the schema, classify it as `MISLEADING_QUERY`.
+- **Time-related Queries:** Don't rephrase time-related information in the user's question.
 
 ### Intent Definitions ###
 
@@ -120,8 +121,10 @@
 {% endif %}
 
 {% if instructions %}
-### INSTRUCTIONS ###
-{{ instructions }}
+### USER INSTRUCTIONS ###
+{% for instruction in instructions %}
+{{ loop.index }}. {{ instruction }}
+{% endfor %}
 {% endif %}
 
 ### USER GUIDE ###
@@ -141,7 +144,6 @@
 {% endif %}
 
 User's current question: {{query}}
-Current Time: {{ current_time }}
 Output Language: {{ language }}
 
 Let's think step by step
@@ -273,7 +275,6 @@ def prompt(
             instructions=instructions,
             configuration=configuration,
         ),
-        current_time=configuration.show_current_time(),
         docs=wren_ai_docs,
     )
 

wren-ai-service/src/pipelines/generation/question_recommendation.py

@@ -1,6 +1,5 @@
 import logging
 import sys
-from datetime import datetime
 from typing import Any
 
 import orjson
@@ -22,7 +21,6 @@ def prompt(
     mdl: dict,
     previous_questions: list[str],
     language: str,
-    current_date: str,
     max_questions: int,
     max_categories: int,
     prompt_builder: PromptBuilder,
@@ -37,7 +35,6 @@ def prompt(
         models=[] if previous_questions else mdl.get("models", []),
         previous_questions=previous_questions,
         language=language,
-        current_date=current_date,
         max_questions=max_questions,
         max_categories=max_categories,
     )
@@ -222,8 +219,6 @@ class QuestionResult(BaseModel):
 Categories: {{categories}}
 {% endif %}
 
-Current Date: {{current_date}}
-
 Please generate {{max_questions}} insightful questions for each of the {{max_categories}} categories based on the provided data model. Both the questions and category names should be translated into {{language}}{% if user_question %} and be related to the user's question{% endif %}. The output format should maintain the structure but with localized text.
 """
 
@@ -255,7 +250,6 @@ async def run(
         previous_questions: list[str] = [],
         categories: list[str] = [],
         language: str = "en",
-        current_date: str = datetime.now().strftime("%Y-%m-%d %A %H:%M:%S"),
         max_questions: int = 5,
         max_categories: int = 3,
         **_,
@@ -268,7 +262,6 @@ async def run(
                 "previous_questions": previous_questions,
                 "categories": categories,
                 "language": language,
-                "current_date": current_date,
                 "max_questions": max_questions,
                 "max_categories": max_categories,
                 **self._components,
@@ -286,7 +279,6 @@ async def run(
         previous_questions=[],
         categories=[],
         language="en",
-        current_date=datetime.now().strftime("%Y-%m-%d %A %H:%M:%S"),
         max_questions=5,
         max_categories=3,
     )

wren-ai-service/src/pipelines/generation/sql_generation.py

@@ -13,7 +13,9 @@
 from src.pipelines.generation.utils.sql import (
     SQL_GENERATION_MODEL_KWARGS,
     SQLGenPostProcessor,
+    calculated_field_instructions,
     construct_instructions,
+    metric_instructions,
     sql_generation_system_prompt,
 )
 from src.pipelines.retrieval.sql_functions import SqlFunction
@@ -28,9 +30,12 @@
     {{ document }}
 {% endfor %}
 
-{% if instructions %}
-### INSTRUCTIONS ###
-{{ instructions }}
+{% if calculated_field_instructions %}
+{{ calculated_field_instructions }}
+{% endif %}
+
+{% if metric_instructions %}
+{{ metric_instructions }}
 {% endif %}
 
 {% if sql_functions %}
@@ -50,9 +55,15 @@
 {% endfor %}
 {% endif %}
 
+{% if instructions %}
+### USER INSTRUCTIONS ###
+{% for instruction in instructions %}
+{{ loop.index }}. {{ instruction }}
+{% endfor %}
+{% endif %}
+
 ### QUESTION ###
 User's Question: {{ query }}
-Current Time: {{ current_time }}
 
 {% if sql_generation_reasoning %}
 ### REASONING PLAN ###
@@ -83,12 +94,13 @@ def prompt(
         sql_generation_reasoning=sql_generation_reasoning,
         instructions=construct_instructions(
             configuration,
-            has_calculated_field,
-            has_metric,
             instructions,
         ),
+        calculated_field_instructions=calculated_field_instructions
+        if has_calculated_field
+        else "",
+        metric_instructions=metric_instructions if has_metric else "",
         sql_samples=sql_samples,
-        current_time=configuration.show_current_time(),
         sql_functions=sql_functions,
     )
 

wren-ai-service/src/pipelines/generation/sql_generation_reasoning.py

@@ -10,33 +10,15 @@
 
 from src.core.pipeline import BasicPipeline
 from src.core.provider import LLMProvider
-from src.pipelines.generation.utils.sql import construct_instructions
+from src.pipelines.generation.utils.sql import (
+    construct_instructions,
+    sql_generation_reasoning_system_prompt,
+)
 from src.web.v1.services import Configuration
 
 logger = logging.getLogger("wren-ai-service")
 
 
-sql_generation_reasoning_system_prompt = """
-### TASK ###
-You are a helpful data analyst who is great at thinking deeply and reasoning about the user's question and the database schema, and you provide a step-by-step reasoning plan in order to answer the user's question.
-
-### INSTRUCTIONS ###
-1. Think deeply and reason about the user's question and the database schema.
-2. Give a step by step reasoning plan in order to answer user's question.
-3. The reasoning plan should be in the language same as the language user provided in the input.
-4. Make sure to consider the current time provided in the input if the user's question is related to the date/time.
-5. Don't include SQL in the reasoning plan.
-6. Each step in the reasoning plan must start with a number, a title(in bold format in markdown), and a reasoning for the step.
-7. If SQL SAMPLES section is provided, make sure to consider them in the reasoning plan.
-8. If INSTRUCTIONS section is provided, please follow them strictly.
-9. Do not include ```markdown or ``` in the answer.
-10. A table name in the reasoning plan must be in this format: `table: <table_name>`.
-11. A column name in the reasoning plan must be in this format: `column: <table_name>.<column_name>`.
-
-### FINAL ANSWER FORMAT ###
-The final answer must be a reasoning plan in plain Markdown string format
-"""
-
 sql_generation_reasoning_user_prompt_template = """
 ### DATABASE SCHEMA ###
 {% for document in documents %}
@@ -54,13 +36,14 @@
 {% endif %}
 
 {% if instructions %}
-### INSTRUCTIONS ###
-{{ instructions }}
+### USER INSTRUCTIONS ###
+{% for instruction in instructions %}
+{{ loop.index }}. {{ instruction }}
+{% endfor %}
 {% endif %}
 
 ### QUESTION ###
 User's Question: {{ query }}
-Current Time: {{ current_time }}
 Language: {{ language }}
 
 Let's think step by step.
@@ -85,7 +68,6 @@ def prompt(
             instructions=instructions,
             configuration=configuration,
         ),
-        current_time=configuration.show_current_time(),
         language=configuration.language,
     )