GitLab

Added OLLAMA_CUSTOM_CUDA_DEFS and OLLAMA_CUSTOM_ROCM_DEFS instead of OLLAMA_CUSTOM_GPU_DEFS

* parse wide argv characters on windows

* cleanup

* move cleanup to end of `main`

TODO: update padding() to _only_ returning the padding

* terminate subprocess if receiving `SIGINT` or `SIGTERM` signals while model is loading

* use `unload` in signal handler

During testing, we're seeing some models take over 3 minutes.

This commit introduces a more friendly way to build Ollama dependencies
and the binary without abusing `go generate` and removing the
unnecessary extra steps it brings with it.

This script also provides nicer feedback to the user about what is
happening during the build process.

At the end, it prints a helpful message to the user about what to do
next (e.g. run the new local Ollama).

update generate scripts with new `LLAMA_CUDA` variable, set `HIP_PLATFORM` to avoid compiler errors (#3528)

The subprocess change moved the build directory
arm64 builds weren't setting cross-compilation flags when building on x86

This should resolve a number of memory leak and stability defects by allowing
us to isolate llama.cpp in a separate process and shutdown when idle, and
gracefully restart if it has problems.  This also serves as a first step to be
able to run multiple copies to support multiple models concurrently.

count each layer independently when deciding gpu offloading

Co-authored-by: Michael Yang <mxyng@pm.me>