-
Notifications
You must be signed in to change notification settings - Fork 323
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Android App Inference Error #6284
Comments
Hi @j0h0k0i0m , can you share the example prompt you are using that can reproduce this crash? Thanks! |
Hi! @Riandy. Thank you for asking for more details about the issue. Here's the prompt I used. system prompt : You are a helpful, respectful and honest assistant. All answers must be in Korean. This input is an example. The app crashes while generating output if the input length is similar or longer. The device currently being used for testing is Galaxy S24 Ultra. The following is the complete error log from the point after input is provided. Error Log
|
The same situation occurs in Japanese and Chinese |
I am looking into this issue and have a fix in https://github.com/pytorch/executorch/pull/6317/files. Seems that we can't convert tokens to partial UTF-16 (or modified UTF-8) characters. |
Hi @j0h0k0i0m @JamePeng , Fix is in #6317 , can you try rebuilding the latest aar package from main, and then try again? Thanks! |
log.txt |
Hi @JamePeng Just want to do a sanity check, could you please run it through adb shell binary
(basically follow .ci/scripts/build_llama_android.sh) |
model : Llama 3.2 1B (without any quantization)
I ran the app, loaded the model, and entered the input, but the following error appears in the middle of the output.
It seems that the issue is occurring because the input language is Korean. Could you advise which part must be modified to resolve this error? I have confirmed no issues when the input language is English.
Tasks
The text was updated successfully, but these errors were encountered: