See https://arxiv.org/abs/2407.07726 for a description of the model.
Because PaliGemma operates as a prefix-LM on the image+prompt, add support for that.
PiperOrigin-RevId: 677841119
This changes the command line flags, but the default value retains the previous behavior.
Also add a CreateGemma helper to enable extra args without interface changes.
PiperOrigin-RevId: 641266411
GEMMA_WEIGHT_T is indeed the correct flag for the C++ compiler,
but the readme references CMake, and there the correct flag name is WEIGHT_TYPE.
PiperOrigin-RevId: 641170380
Also implement support for some model variations:
- Local attention.
- Add support for biases.
- Use RoPE only on half vectors.
- Support different order of QKV weights.
Co-authored-by: Andrey Mikhaylov <amik@google.com>
Co-authored-by: Martin Bruse <zondolfin@gmail.com>
Co-authored-by: Zoltan Szabadka <szabadka@google.com>
--
19694e1f2e by Silvio Traversaro <silvio@traversaro.it>:
Do not pass explicitly -O2 flag to compiler in Release build
COPYBARA_INTEGRATE_REVIEW=https://github.com/google/gemma.cpp/pull/3 from traversaro:patch-1 19694e1f2e
PiperOrigin-RevId: 610096914
Using a `CMakePresets.json` file makes it much easier to manage
several alternate build configurations, such as the "ClangCL"
build for Windows. This also makes it easier for tools like
VSCode to run CMake-based builds.
Discussed using dev branch instead of main branch for development.
We could revisit this once CI is hardened to test end-to-end correctness.
PiperOrigin-RevId: 609500897
Daniel Keysers
Jan Wassenberg
Paul Chang
zond
Luca Versari
austinvhuang
Copybara-Service
Omar Sanseviero
Dan Zheng
Naoki Kishida
Silvio Traversaro
Jan Wassenberg
The gemma.cpp Authors
David Coles
Austin Huang