GitHub Issue 模板 (建议标题及正文)
Title:
[Bug]: Gemma-4-E4B-it Initialization Failed - LiteRT Engine Error on Mobile
Description:
Hi development team,
I encountered a consistent crash when attempting to run the Gemma-4-E4B-it model within the Google AI Edge Gallery (Beta) app. The engine fails to initialize immediately after selecting the model.
1. Environment:
App: Google AI Edge Gallery (Early Access)
Model: Gemma-4-E4B-it
Update Date: April 1, 2026
2. Error Message:
The app throws a detailed internal error dialog:
Failed to create engine: INTERNAL: ERROR: [third_party/odml/litert_lm/runtime/executor/llm_litert_compiled_model_executor.cc:1925] ∟ ERROR: [./third_party/odml/litert/litert/cc/litert_compiled_model.h:836]
3. Observed Behavior:
The model selection starts the loading process, but instead of entering the chat interface, it triggers the “Failed to create engine” popup. This suggests a failure during the model compilation or NPU/GPU delegate assignment phase in the LiteRT runtime.
4. Attachments:
(Here, please upload the two screenshots: 16268.jpg and 16234.jpg)
Note to User: In GitHub’s editor, you can just drag and drop the files here.
5. Additional Context:
I previously reported this on the Google Play Store, but the automated moderation system incorrectly flagged the technical error logs as “gibberish.” Providing this here to ensure the stack trace reaches the engineers.
Hi @Guang_Long,Thanks for reaching out and testing the model!
I noticed you also opened an issue over on the LiteRT GitHub repository; thank you for doing so. Since this specific error is tied to the on-device inference engine rather than the Gemma model weights, GitHub is exactly the right place for it.
I see that the LiteRT engineering team has already replied to your thread and is waiting for a few hardware details to help troubleshoot the compilation crash. Please continue the conversation with them there so they can help get this sorted out for you!
If you run into any behavior, prompting, or quality issues specifically related to the Gemma 4 model down the line, please feel free to open a new thread here. We’d be happy to help.
Thank you!
