π Bug Description
When testing LLMEngineService.generateStreamingResponse(), we found the nativeStreamingInference() is not returned normally.
We put a breakpoint at line #453 and didn't see the stop. That will make the resultFuture.complete() miss the invocation.
π― Steps to Reproduce
- Setup NPU configuration in AppConstants.java (ex. BACKEND_DEFAULT, DEFAULT_LLM_MODEL)
- Launch ChatActivit and enter prompt message: "δ½ ζ―θͺ°"
π± Test Device
Device: VIVO X200s
OS Version: Android 15
Screen Resolution: 2800 Γ 1260 pixels