GPUs handled the dependency path properly, but CPU runners didn't which
results in missing vc redist libraries on systems where the user didn't
already have it installed from some other app.
* Optimize container images for startup
This change adjusts how to handle runner payloads to support
container builds where we keep them extracted in the filesystem.
This makes it easier to optimize the cpu/cuda vs cpu/rocm images for
size, and should result in faster startup times for container images.
* Refactor payload logic and add buildx support for faster builds
* Move payloads around
* Review comments
* Converge to buildx based helper scripts
* Use docker buildx action for release
This adds back a check which was lost many releases back to verify /dev/kfd permissions
which when lacking, can lead to confusing failure modes of:
"rocBLAS error: Could not initialize Tensile host: No devices found"
This implementation does not hard fail the serve command but instead will fall back to CPU
with an error log. In the future we can include this in the GPU discovery UX to show
detected but unsupported devices we discovered.
It looks like driver 525 (aka, cuda driver 12.0) has problems with the cuda v12 library
we compile against, so run v11 on those older drivers if detected.
During rebasing, the ordering was inverted causing the cuda version
selection logic to break, with driver version being evaluated as zero
incorrectly causing a downgrade to v11.
This adjusts linux to follow a similar model to windows with a discrete archive
(zip/tgz) to cary the primary executable, and dependent libraries. Runners are
still carried as payloads inside the main binary
Darwin retain the payload model where the go binary is fully self contained.
If the system has multiple numa nodes, enable numa support in llama.cpp
If we detect numactl in the path, use that, else use the basic "distribute" mode.
The v5 hip library returns unsupported GPUs which wont enumerate at
inference time in the runner so this makes sure we align discovery. The
gfx906 cards are no longer supported so we shouldn't compile with that
GPU type as it wont enumerate at runtime.
* llm: avoid loading model if system memory is too small
* update log
* Instrument swap free space
On linux and windows, expose how much swap space is available
so we can take that into consideration when scheduling models
* use `systemSwapFreeMemory` in check
---------
Co-authored-by: Daniel Hiltgen <daniel@ollama.com>
This also adjusts our algorithm to favor our bundled ROCm.
I've confirmed VRAM reporting still doesn't work properly so we
can't yet enable concurrency by default.
This adds logic to detect skew between the driver and
management library which can be attributed to OS overhead
and records that so we can adjust subsequent management
library free VRAM updates and avoid OOM scenarios.
Refine the way we log GPU discovery to improve the non-debug
output, and report more actionable log messages when possible
to help users troubleshoot on their own.
Until ROCm v6.2 ships, we wont be able to get accurate free memory
reporting on windows, which makes automatic concurrency too risky.
Users can still opt-in but will need to pay attention to model sizes otherwise they may thrash/page VRAM or cause OOM crashes.
All other platforms and GPUs have accurate VRAM reporting wired
up now, so we can turn on concurrency by default.
We update the PATH on windows to get the CLI mapped, but this has
an unintended side effect of causing other apps that may use our bundled
DLLs to get terminated when we upgrade.
