xThr45hx/Gemma3-1B-IT-Tensor-G4-NPU
Text Generation • Updated • 11 • 1
Tensor G4 Edge TPU — On-device Gemma + RAG (LiteRT-LM)
int4 AOT Gemma 3 1B + on-device EmbeddingGemma RAG (Termux engine + MCP) on the Pixel 9 Tensor G4 Edge TPU via LiteRT-LM.
Note Gemma 3 1B-IT for the G4 NPU: int4 per-channel AOT (~100% NPU, ~half the G5 q8 bundle) + two plug-and-play JIT bundles + the rank-2 FC surgery.
xThr45hx/EmbeddingGemma-300M-Tensor-G4-NPU
Feature Extraction • UpdatedNote EmbeddingGemma-300M on the G4 NPU + fully on-device Termux RAG: native engine runner, vector index, MCP server (search_context).