GitLab

This refines where we extract the LLM libraries to by adding a new
OLLAMA_HOME env var, that defaults to `~/.ollama` The logic was already
idempotenent, so this should speed up startups after the first time a
new release is deployed.  It also cleans up after itself.

We now build only a single ROCm version (latest major) on both windows
and linux.  Given the large size of ROCms tensor files, we split the
dependency out.  It's bundled into the installer on windows, and a
separate download on windows.  The linux install script is now smart and
detects the presence of AMD GPUs and looks to see if rocm v6 is already
present, and if not, then downloads our dependency tar file.

For Linux discovery, we now use sysfs and check each GPU against what
ROCm supports so we can degrade to CPU gracefully instead of having
llama.cpp+rocm assert/crash on us.  For Windows, we now use go's windows
dynamic library loading logic to access the amdhip64.dll APIs to query
the GPU information.

Make sure that when a shutdown signal comes, we shutdown quickly instead
of waiting for a potentially long exchange to wrap up.

* Fix clearing kv cache between requests with the same prompt

* fix powershell script

This wires up logging in llama.cpp to always go to stderr, and also
turns up logging if OLLAMA_DEBUG is set.

A few obvious levels were adjusted, but generally everything mapped to "info" level.

This also refines the build process for the ext_server build.

- prompt cache causes inferance to hang after some time

Make sure we're building an x86 ext_server lib when cross-compiling

This switches darwin to dynamic loading, and refactors the code now that no
static linking of the library is used on any platform

This reduces the built-in linux version to not use any vector extensions
which enables the resulting builds to run under Rosetta on MacOS in
Docker.  Then at runtime it checks for the actual CPU vector
extensions and loads the best CPU library available

* select layers based on estimated model memory usage

* always account for scratch vram

* dont load +1 layers

* better estmation for graph alloc

* Update gpu/gpu_darwin.go
Co-authored-by: Bruce MacDonald <brucewmacdonald@gmail.com>

* Update llm/llm.go
Co-authored-by: Bruce MacDonald <brucewmacdonald@gmail.com>

* Update llm/llm.go

* add overhead for cuda memory

* Update llm/llm.go
Co-authored-by: Bruce MacDonald <brucewmacdonald@gmail.com>

* fix build error on linux

* address comments

---------
Co-authored-by: Bruce MacDonald <brucewmacdonald@gmail.com>

Refactor where we store build outputs, and support a fully dynamic loading
model on windows so the base executable has no special dependencies thus
doesn't require a special PATH.

This changes the model for llama.cpp inclusion so we're not applying a patch,
but instead have the C++ code directly in the ollama tree, which should make it
easier to refine and update over time.

The default thread count logic was broken and resulted in 2x the number
of threads as it should on a hyperthreading CPU
resulting in thrashing and poor performance.

The windows native setup still needs some more work, but this gets it building
again and if you set the PATH properly, you can run the resulting exe on a cuda system.

This switches the default llama.cpp to be CPU based, and builds the GPU variants
as dynamically loaded libraries which we can select at runtime.

This also bumps the ROCm library to version 6 given 5.7 builds don't work
on the latest ROCm library that just shipped.

If someone checks out the ollama repo and doesn't install the CUDA
library, this will ensure they can build a CPU only version

Run the server.cpp directly inside the Go runtime via cgo
while retaining the LLM Go abstractions.