Skip to content

armgit / Edge0

Public Imported from GitHub
0

Run 35B MoE models in ~2.5 GB of RAM. Weights stream from SSD; a trained router predicts loads a step ahead. Mac, iPhone, Android, Windows supported

Python
1.1 KB
Raw Blame
NOTICE - edge0 / android
================================

Licensed under Apache-2.0 (see the LICENSE at the monorepo root). Besides our
own code, this project contains or references the following third-party material:

1. llama.cpp - MIT (commit 7ab4ee7, release b11100, pinned)
   Location: ../vendor/llama.cpp (pristine upstream checkout, never patched in
   place). Built with ../patches/llama.cpp/{common,android} replayed (20
   hook-point bands) into the shipped shared libraries (libllama, libggml*).
   ggml (Apache-2.0) is developed inside the same upstream tree. Not vendored
   as files; built from source by the consumer (tools/llama/build_vendor_libs.sh).

2. Gradle/Maven dependencies (Apache-2.0 / MIT / BSD)
   Kotlin, Jetpack Compose, AndroidX (Activity, Room, DataStore, Lifecycle),
   kotlinx-serialization, mikepenz multiplatform-markdown-renderer-m3 (MIT),
   JUnit. Resolved at build time; each ships its own license text.

3. Model assets - Edge0/Edge0-8B-A1B-preview and Edge0/Edge0-35B-A3B-preview on
   Hugging Face; terms per the model cards. Not part of the source distribution.