Search results

[LiteRT.js / WebGPU] CompiledModel produces correct output only on its first inference on Adreno 8xx (Galaxy Z Fold 7, Chrome 149)

Status: Open.

#8065 In google-ai-edge/LiteRT;

· Harunene opened on Jun 12, 2026

[Google Tensor] Generative decoder LLMs (Gemma-3-1B, Gemma-4, Qwen2.5) fail to compile to Tensor_G5 with INTERNAL; vision compiles in the same env

Status: Open.

#8062 In google-ai-edge/LiteRT;

· mlnomadpy opened on Jun 12, 2026

[LiteRT] CompiledModel.create() fails to trigger on-device JIT for float models, and host-side AOT compiler ignores --soc_model (hardcoded to v79)

type:bugBugBug

Status: Open.

#8037 In google-ai-edge/LiteRT;

· akshayOnform opened on Jun 11, 2026

Rust segmentation example fails to build and run on macOS (arm64)

status:awaiting LiteRTerinformation is sufficient and the issue is escalatedinformation is sufficient and the issue is escalated

status:awaiting PR reviewawaiting a PR reviewawaiting a PR review

type:bugBugBug

Status: Open.

#7997 In google-ai-edge/LiteRT;

· tallamjr opened on Jun 10, 2026

Expose Custom OP Dispatcher to Python API

type:featureFor feature requestsFor feature requests

Status: Open.

#7996 In google-ai-edge/LiteRT;

· tackyattack opened on Jun 10, 2026

EfficientDet fails on Android GPU (S20/S23) with CompiledModel

type:bugBugBug

Status: Open.

#7936 In google-ai-edge/LiteRT;

· ekrynicki-bah opened on Jun 8, 2026

CompiledModel: NPU silently falls back to CPU with no diagnostic for custom models (S23 Ultra / Pixel 9 Pro)

type:bugBugBug

Status: Open.

#7891 In google-ai-edge/LiteRT;

· akshayOnform opened on Jun 5, 2026

[Intel-OpenVINO] The openvino_compiler_plugin_test and dispatch_api_openvino_test targets cannot be built

status:awaiting LiteRTerinformation is sufficient and the issue is escalatedinformation is sufficient and the issue is escalated

status:awaiting PR reviewawaiting a PR reviewawaiting a PR review

type:bugBugBug

Status: Open.

#7878 In google-ai-edge/LiteRT;

· daijh opened on Jun 5, 2026

LiteRT Next CompiledModel + Accelerator.NPU (Qualcomm HTP) ~6x slower than the Interpreter + QNN TFLite Delegate path for the same INT8 model; QualcommOptions have no measurable effect

status:awaiting user responseawaiting user responseawaiting user response

type:supportSupport IssuesSupport Issues

Status: Open.

#7858 In google-ai-edge/LiteRT;

· wnbotoo opened on Jun 4, 2026

[Feature Request] Native support for 1-bit and Ternary (1.58-bit) quantization

type:featureFor feature requestsFor feature requests

Status: Open.

#7713 In google-ai-edge/LiteRT;

· biaji opened on Jun 1, 2026

EmbeddingGemma 300M returns all-zero vectors on Mac GPU unless precision is forced to FP32

status:awaiting LiteRTerinformation is sufficient and the issue is escalatedinformation is sufficient and the issue is escalated

type:bugBugBug

Status: Open.

#7693 In google-ai-edge/LiteRT;

· TheMrCodes opened on May 30, 2026

Model compilation failed for Pixel 10 Pro on Ubuntu 20.04 via Google Tensor SDK Beta for NPU inference

type:bugBugBug

Status: Open.

#7629 In google-ai-edge/LiteRT;

· AlexiAlp opened on May 27, 2026