Feature request: Improve long-chat code generation and allow continuing incomplete code output in Qwen Studio

起始日期 | Start Date

No response

实现PR | Implementation PR

No response

相关Issues | Reference Issues

No response

摘要 | Summary

When a conversation in Qwen Studio becomes long, code generation may stop before the code is complete. This results in incomplete code blocks and interrupts coding workflows.

It would be useful to improve long-chat stability and provide a way to continue or resume incomplete code generation, especially for large code outputs.

基本示例 | Basic Example

Example scenario:

1. Open Qwen Studio.
2. Have a long coding conversation with accumulated context.
3. Ask the model to generate a large code file, class, or function.
4. The model starts writing the code but stops before the code block is finished.
5. The output becomes incomplete, for example:
   • a function is missing its closing part;
   • a class or file is not fully generated;
   • the code block stops in the middle.

Expected improvement:

• Add a “Continue generating” option for incomplete responses.
• Automatically continue or prompt continuation when code output is cut off.
• Preserve context so the continued code matches the previous output.
• Improve handling of long conversations to reduce unexpected stopping.

缺陷 | Drawbacks

Possible drawbacks:

• Continuing long outputs may require additional tokens and processing time.
• The continued code may sometimes be inconsistent with the earlier generated code.
• Implementing reliable continuation/resume behavior may require UI and backend changes.
• In very long conversations, preserving full context may still be limited by the model context window.

未解决问题 | Unresolved questions

No response