j4k3@lemmy.world to Google Pixel@lemmy.worldEnglish · 1 year agoIs anyone here a dev that can give a heads up about the Tensor SoC API, is it open source, toolchain?message-squaremessage-square3fedilinkarrow-up18arrow-down10file-text
arrow-up18arrow-down1message-squareIs anyone here a dev that can give a heads up about the Tensor SoC API, is it open source, toolchain?j4k3@lemmy.world to Google Pixel@lemmy.worldEnglish · 1 year agomessage-square3fedilinkfile-text
I’m just curious about the potential of running a small LLM locally on GrapheneOS, maybe some image processing too.
minus-squareAlmightySnoo 🐢🇮🇱🇺🇦@lemmy.worldMlinkfedilinkEnglisharrow-up3·edit-21 year agoI think you may be referring to the Android NN API, which apparently should use the backend targeting Google Tensor SoCs when available. From this commit it seems it should be available on GrapheneOS too: https://github.com/GrapheneOS/device_google_gs201/commit/b60bfdd87550bf20f6cb73234a1a8ed2ecd61b69 (EdgeTPU is actually the ASIC differentiating Tensor SoCs from the rest by allowing fast and low-power inference from a hardware level).
I think you may be referring to the Android NN API, which apparently should use the backend targeting Google Tensor SoCs when available. From this commit it seems it should be available on GrapheneOS too: https://github.com/GrapheneOS/device_google_gs201/commit/b60bfdd87550bf20f6cb73234a1a8ed2ecd61b69 (EdgeTPU is actually the ASIC differentiating Tensor SoCs from the rest by allowing fast and low-power inference from a hardware level).