feat: support suggested_questions_after_answer to be configed (#29254)

Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>

api/.env.example

@@ -633,6 +633,22 @@ SWAGGER_UI_PATH=/swagger-ui.html
 # Set to false to export dataset IDs as plain text for easier cross-environment import
 DSL_EXPORT_ENCRYPT_DATASET_ID=true
 
+# Suggested Questions After Answer Configuration
+# These environment variables allow customization of the suggested questions feature
+#
+# Custom prompt for generating suggested questions (optional)
+# If not set, uses the default prompt that generates 3 questions under 20 characters each
+# Example: "Please help me predict the five most likely technical follow-up questions a developer would ask. Focus on implementation details, best practices, and architecture considerations. Keep each question between 40-60 characters. Output must be JSON array: [\"question1\",\"question2\",\"question3\",\"question4\",\"question5\"]"
+# SUGGESTED_QUESTIONS_PROMPT=
+
+# Maximum number of tokens for suggested questions generation (default: 256)
+# Adjust this value for longer questions or more questions
+# SUGGESTED_QUESTIONS_MAX_TOKENS=256
+
+# Temperature for suggested questions generation (default: 0.0)
+# Higher values (0.5-1.0) produce more creative questions, lower values (0.0-0.3) produce more focused questions
+# SUGGESTED_QUESTIONS_TEMPERATURE=0
+
 # Tenant isolated task queue configuration
 TENANT_ISOLATED_TASK_CONCURRENCY=1
 

api/core/llm_generator/llm_generator.py

@@ -15,6 +15,8 @@ from core.llm_generator.prompts import (
     LLM_MODIFY_CODE_SYSTEM,
     LLM_MODIFY_PROMPT_SYSTEM,
     PYTHON_CODE_GENERATOR_PROMPT_TEMPLATE,
+    SUGGESTED_QUESTIONS_MAX_TOKENS,
+    SUGGESTED_QUESTIONS_TEMPERATURE,
     SYSTEM_STRUCTURED_OUTPUT_GENERATE,
     WORKFLOW_RULE_CONFIG_PROMPT_GENERATE_TEMPLATE,
 )
@@ -124,7 +126,10 @@ class LLMGenerator:
         try:
             response: LLMResult = model_instance.invoke_llm(
                 prompt_messages=list(prompt_messages),
-                model_parameters={"max_tokens": 256, "temperature": 0},
+                model_parameters={
+                    "max_tokens": SUGGESTED_QUESTIONS_MAX_TOKENS,
+                    "temperature": SUGGESTED_QUESTIONS_TEMPERATURE,
+                },
                 stream=False,
             )
 

api/core/llm_generator/prompts.py

@@ -1,4 +1,6 @@
 # Written by YORKI MINAKO🤡, Edited by Xiaoyi, Edited by yasu-oh
+import os
+
 CONVERSATION_TITLE_PROMPT = """You are asked to generate a concise chat title by decomposing the user’s input into two parts: “Intention” and “Subject”.
 
 1. Detect Input Language
@@ -94,7 +96,8 @@ JAVASCRIPT_CODE_GENERATOR_PROMPT_TEMPLATE = (
 )
 
 
-SUGGESTED_QUESTIONS_AFTER_ANSWER_INSTRUCTION_PROMPT = (
+# Default prompt for suggested questions (can be overridden by environment variable)
+_DEFAULT_SUGGESTED_QUESTIONS_AFTER_ANSWER_PROMPT = (
     "Please help me predict the three most likely questions that human would ask, "
     "and keep each question under 20 characters.\n"
     "MAKE SURE your output is the SAME language as the Assistant's latest response. "
@@ -102,6 +105,15 @@ SUGGESTED_QUESTIONS_AFTER_ANSWER_INSTRUCTION_PROMPT = (
     '["question1","question2","question3"]\n'
 )
 
+# Environment variable override for suggested questions prompt
+SUGGESTED_QUESTIONS_AFTER_ANSWER_INSTRUCTION_PROMPT = os.getenv(
+    "SUGGESTED_QUESTIONS_PROMPT", _DEFAULT_SUGGESTED_QUESTIONS_AFTER_ANSWER_PROMPT
+)
+
+# Configurable LLM parameters for suggested questions (can be overridden by environment variables)
+SUGGESTED_QUESTIONS_MAX_TOKENS = int(os.getenv("SUGGESTED_QUESTIONS_MAX_TOKENS", "256"))
+SUGGESTED_QUESTIONS_TEMPERATURE = float(os.getenv("SUGGESTED_QUESTIONS_TEMPERATURE", "0"))
+
 GENERATOR_QA_PROMPT = (
     "<Task> The user will send a long text. Generate a Question and Answer pairs only using the knowledge"
     " in the long text. Please think step by step."