Stop sequences configured on non-workflow apps are never sent to the model

Start with ModelConfigManager.convert and ModelConfigConverter.convert in core/app/app_config/easy_ui_based_app/model_config, then inspect the affected consumers at core/app/apps/base_app_runner.py:167, core/agent/cot_agent_runner.py:150, and core/agent/fc_agent_runner.py:182. Run or extend the regression test mentioned in the issue and confirm that configured stop sequences reach invoke_llm for non-workflow and agent apps.

Self Checks

• I have read the Contributing Guide and Language Policy.
• This is only for bug report, if you would like to ask a question, please head to Discussions.
• I have searched for existing issues search for existing issues, including closed ones.
• I confirm that I am using English to submit this report.
• [FOR CHINESE USERS] 请务必使用英文提交 Issue,否则会被关闭。谢谢!:)
• Please do not modify this template :) and fill in all the required fields.

Dify version

1.17.0

Cloud or Self Hosted

Self Hosted (Source)

Steps to reproduce

1. Create a Chatbot app (this also affects Agent, Completion and Agent Chat apps).
2. Switch the prompt to expert mode.
3. Open the model parameter panel and add a stop sequence, for example ###.
4. Publish, then send a message that makes the model produce ### mid answer.
5. The model keeps generating past ###.

The saved app model config is correct, the loss happens during runtime conversion. ModelConfigManager.convert moves stop out of completion_params and onto ModelConfigEntity.stop, but ModelConfigConverter.convert still looks for stop inside model_config.parameters, where it no longer exists, and never reads model_config.stop. ModelConfigWithCredentialsEntity.stop is therefore always empty, and ModelConfigEntity.stop has no readers anywhere in the codebase.

from types import SimpleNamespace
from unittest.mock import MagicMock, patch
from core.app.app_config.easy_ui_based_app.model_config.converter import ModelConfigConverter
from core.app.app_config.easy_ui_based_app.model_config.manager import ModelConfigManager

M = "core.app.app_config.easy_ui_based_app.model_config.converter"
saved = {"model": {"provider": "openai", "name": "gpt-4", "mode": "chat",
                   "completion_params": {"temperature": 0.5, "stop": ["###"]}}}
entity = ModelConfigManager.convert(saved)
print("ModelConfigEntity.stop      ->", entity.stop)

with (patch(f"{M}.create_plugin_provider_manager"),
      patch(f"{M}.ModelConfigWithCredentialsEntity", side_effect=lambda **kw: SimpleNamespace(**kw))):
    runtime = ModelConfigConverter.convert(MagicMock(tenant_id="t", model=entity))
print("stop actually sent to model ->", runtime.stop)

ModelConfigEntity.stop      -> ['###']
stop actually sent to model -> []

Affected consumers are core/app/apps/base_app_runner.py:167 for the advanced prompt path and core/agent/cot_agent_runner.py:150 and core/agent/fc_agent_runner.py:182 for agent apps.

I have a fix ready with a regression test and would be happy to be assigned to this issue.

✔️ Expected Behavior

The stop sequences saved in the model configuration reach invoke_llm, so generation stops at the configured sequence.

❌ Actual Behavior

ModelConfigWithCredentialsEntity.stop is always an empty list, so the model is invoked with no stop sequences and generation runs past them.