Dataset Viewer

The dataset viewer is not available because its heuristics could not detect any supported data files. You can try uploading some data files, or configuring the data files location manually.

MobileTransformers — build artifacts

This is not a dataset. It is the set of build inputs that cannot live in git, published here so that a git clone of MobileTransformers can provision itself.

Nothing here is downloaded by the Android app or by any model package. These files are consumed by one script, at development time:

scripts/fetch_native_deps.sh              # the natives bundle — required to build the Android SDK
TRAINING=1 scripts/fetch_native_deps.sh   # + the ORT-training wheel, for exporting a training stage
SYMBOLS=1  scripts/fetch_native_deps.sh   # + unstripped binaries, for symbolicating a native crash

That script reads third_party/android/manifest.json, downloads what is missing, checks the archive sha256, unpacks it, and then checks every unpacked file's sha256 individually. Both halves matter: the archive hash proves the download, the per-file hashes prove the unpack, and a half-populated jniLibs/ is the failure mode that produces a linker error naming a symbol rather than a file.

Contents

file size needed for
mobiletransformers-natives-0.2.0-arm64-v8a.tar.zst 63 MB Required to build the Android SDK: ONNX Runtime (training build), the GenAI engine, the tokenizer static libs, and vendored headers.
onnxruntime_training-1.23.0+cpu-cp312-cp312-linux_x86_64.whl 632 MB Only to export a training stage. Source-built, cp312/linux_x86_64 only — it is not on PyPI.
mobiletransformers-natives-0.2.0-arm64-v8a-debug-symbols.tar.zst 261 MB Optional. The unstripped originals of the shipped .so files. Android's build strips them at packaging, so these cost nothing at runtime and are the only way to read a native stack trace.

Every sha256, size, provenance and role is recorded in third_party/android/manifest.json and third_party/onnxruntime/manifest.json. Verify by hand with sha256sum if you prefer; the fetch script does it for you either way.

Licensing

These are builds of third-party projects — ONNX Runtime, onnxruntime-genai, tokenizers-cpp, protobuf — each under its own upstream licence. See THIRD_PARTY_NOTICES.md. The license: other tag above reflects that this repo is a mixed bundle of upstream artifacts rather than a single licensed work.

arm64-v8a only

There is no x86_64 build, so the SDK does not run on a standard Android emulator. libonnxruntime.so and the tokenizer archives were never built for it; restoring x86_64 means building ONNX Runtime Training and tokenizers-cpp for that ABI first.

Downloads last month
49