opensource/dify
2
0
Fork 0
mirror of https://github.com/langgenius/dify.git synced 2026-05-09 12:59:18 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix: suggest questions more max_tokens (#35533)

Browse Source
This commit is contained in:
非法操作 2026-04-24 16:02:30 +08:00 committed by fatelei
parent 457b4a5e48
commit 34c1caec48
No known key found for this signature in database
GPG Key ID: 2F91DA05646F4EED
4 changed files with 14 additions and 16 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
api/core/llm_generator/llm_generator.py
View File

@ -13,8 +13,6 @@ from core.llm_generator.output_parser.rule_config_generator import RuleConfigGen
from core.llm_generator.output_parser.suggested_questions_after_answer import SuggestedQuestionsAfterAnswerOutputParser
from core.llm_generator.prompts import (
CONVERSATION_TITLE_PROMPT,
DEFAULT_SUGGESTED_QUESTIONS_MAX_TOKENS,
DEFAULT_SUGGESTED_QUESTIONS_TEMPERATURE,
GENERATOR_QA_PROMPT,
JAVASCRIPT_CODE_GENERATOR_PROMPT_TEMPLATE,
LLM_MODIFY_CODE_SYSTEM,
@ -217,8 +215,8 @@ class LLMGenerator:
else:
# Default-model generation keeps the built-in suggested-questions tuning.
model_parameters = {
"max_tokens": DEFAULT_SUGGESTED_QUESTIONS_MAX_TOKENS,
"temperature": DEFAULT_SUGGESTED_QUESTIONS_TEMPERATURE,
"max_tokens": 2560,
"temperature": 0.0,
}
stop = []

9
api/core/llm_generator/output_parser/suggested_questions_after_answer.py
View File

@ -10,7 +10,14 @@ logger = logging.getLogger(__name__)
class SuggestedQuestionsAfterAnswerOutputParser:
def __init__(self, instruction_prompt: str | None = None) -> None:
self._instruction_prompt = instruction_prompt or DEFAULT_SUGGESTED_QUESTIONS_AFTER_ANSWER_INSTRUCTION_PROMPT
self._instruction_prompt = self._build_instruction_prompt(instruction_prompt)
@staticmethod
def _build_instruction_prompt(instruction_prompt: str | None) -> str:
if not instruction_prompt or not instruction_prompt.strip():
return DEFAULT_SUGGESTED_QUESTIONS_AFTER_ANSWER_INSTRUCTION_PROMPT
return f'{instruction_prompt}\nYou must output a JSON array like ["question1", "question2", "question3"].'
def get_format_instructions(self) -> str:
return self._instruction_prompt

3
api/core/llm_generator/prompts.py
View File

@ -104,9 +104,6 @@ DEFAULT_SUGGESTED_QUESTIONS_AFTER_ANSWER_INSTRUCTION_PROMPT = (
'["question1","question2","question3"]\n'
)
DEFAULT_SUGGESTED_QUESTIONS_MAX_TOKENS = 256
DEFAULT_SUGGESTED_QUESTIONS_TEMPERATURE = 0.0
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."

12
api/tests/unit_tests/core/llm_generator/test_llm_generator.py
View File

@ -6,10 +6,6 @@ import pytest
from core.app.app_config.entities import ModelConfig
from core.llm_generator.entities import RuleCodeGeneratePayload, RuleGeneratePayload, RuleStructuredOutputPayload
from core.llm_generator.llm_generator import LLMGenerator
from core.llm_generator.prompts import (
DEFAULT_SUGGESTED_QUESTIONS_MAX_TOKENS,
DEFAULT_SUGGESTED_QUESTIONS_TEMPERATURE,
)
from graphon.model_runtime.entities.llm_entities import LLMMode, LLMResult
from graphon.model_runtime.entities.model_entities import ModelType
from graphon.model_runtime.errors.invoke import InvokeAuthorizationError, InvokeError
@ -102,8 +98,8 @@ class TestLLMGenerator:
assert len(questions) == 2
assert questions[0] == "Question 1?"
assert mock_model_instance.invoke_llm.call_args.kwargs["model_parameters"] == {
"max_tokens": DEFAULT_SUGGESTED_QUESTIONS_MAX_TOKENS,
"temperature": DEFAULT_SUGGESTED_QUESTIONS_TEMPERATURE,
"max_tokens": 2560,
"temperature": 0.0,
}
def test_generate_suggested_questions_after_answer_auth_error(self, mock_model_instance):
@ -181,8 +177,8 @@ class TestLLMGenerator:
model_type=ModelType.LLM,
)
assert default_model_instance.invoke_llm.call_args.kwargs["model_parameters"] == {
"max_tokens": DEFAULT_SUGGESTED_QUESTIONS_MAX_TOKENS,
"temperature": DEFAULT_SUGGESTED_QUESTIONS_TEMPERATURE,
"max_tokens": 2560,
"temperature": 0.0,
}
assert default_model_instance.invoke_llm.call_args.kwargs["stop"] == []