GitLab

* test: harden scheduler tests

This removes reschedDelay which was stale code, and adds
a new configurable timeout for the waitForVRAMRecovery so
tests can now set the timeout to be very short to avoid the
scheduler getting stuck and hitting a test timeout.

* test: tune tests for partial loads

Give stress tests more time when the model is split between CPU/GPU

This revamps how we discover GPUs in the system by leveraging the Ollama
runner.  This should eliminate inconsistency between our GPU discovery and the
runners capabilities at runtime, particularly for cases where we try to filter
out unsupported GPUs.  Now the runner does that implicitly based on the actual
device list.  In some cases free VRAM reporting can be unreliable which can
leaad to scheduling mistakes, so this also includes a patch to leverage more
reliable VRAM reporting libraries if available.

Automatic workarounds have been removed as only one GPU leveraged this, which
is now documented. This GPU will soon fall off the support matrix with the next
ROCm bump.

Additional cleanup of the scheduler and discovery packages can be done in the
future once we have switched on the new memory management code, and removed
support for the llama runner.

This changes the memory allocation strategy from upfront estimation to
tracking actual allocations done by the engine and reacting to that. The
goal is avoid issues caused by both under-estimation (crashing) and
over-estimation (low performance due to under-utilized GPUs).

It is currently opt-in and can be enabled for models running on the
Ollama engine by setting OLLAMA_NEW_ESTIMATES=1. Behavior in other
cases is unchanged and will continue to use the existing estimates.

* Reapply "feat: incremental gguf parser (#10822)" (#11114)

This reverts commit a6e64fbd.

* fix older ggufs

This reverts commit 6b04cad7.

* incremental gguf parser
* gguf: update test to not rely on gguf on disc
* re-use existing create gguf
* read capabilities from gguf kv
* kv exists
* update tests
* s/doneFunc/successFunc/g
* new buffered reader

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

* Move quantization logic to GGML via new backend

This moves the model aware logic to Go code and calls GGMLs quantization code for model creation.

* Remove "add model quantizations"

This is no longer needed now that quantization is implemented in Go+GGML code directly.

This enhances our logging in the scheduler.  The initial "waiting for server" log
no longer claims an initial error state (now "not responding" which better reflects
the actual state).  Runners now have slog wiring to report more details about the
runner, including PID.

This reverts commit 424f6486.

* increase default context length to 4096

We lower the default numParallel from 4 to 2 and use these "savings" to
double the default context length from 2048 to 4096.

We're memory neutral in cases when we previously would've used
numParallel == 4, but we add the following mitigation to handle some
cases where we would have previously fallen back to 1x2048 due to low
VRAM: we decide between 2048 and 4096 using a runtime check, choosing
2048 if we're on a one GPU system with total VRAM of <= 4 GB. We
purposefully don't check the available VRAM because we don't want the
context window size to change unexpectedly based on the available VRAM.

We plan on making the default even larger, but this is a relatively
low-risk change we can make to quickly double it.

* fix tests

add an explicit context length so they don't get truncated. The code
that converts -1 from being a signal for doing a runtime check isn't
running as part of these tests.

* tweak small gpu message

* clarify context length default

also make it actually show up in `ollama serve --help`

Fix flake failures on windows

feat: add new Ollama engine using ggml through cgo

This change introduces a new way to run pretrained models. It introduces 3 high level interfaces and a bunch of smaller helper interfaces to facilitate this.

- `model.Model` defines the interface for a model architecture. Models such as `llama` and `mllama`, which are provided as examples, can implement the model's forward propagation in the `Forward` method. This method will be called to generate completions. This interface can be found in `model/model.go`
- `ml.Backend` defines the interface for a backend tensor library, in this case `ggml`. Among other things, a Backend is responsible for loading a pretrained model into hardware (GPU, CPU, etc) and providing an interface for Models to access loaded tensors. This interface can be found in `ml/backend.go`
- `ml.Tensor` defines the interface for a tensor and tensor operations

This is the first implementation of the new engine. Follow up PRs will implement more features:

- non-greedy sampling (#8410)
- integration with Ollama and KV caching (#8301)
- more model support (#9080) with more coming soon
Co-authored-by: Bruce MacDonald <brucewmacdonald@gmail.com>

Cleaning up go package naming

* Unified arm/x86 windows installer

This adjusts the installer payloads to be architecture aware so we can cary
both amd64 and arm64 binaries in the installer, and install only the applicable
architecture at install time.

* Include arm64 in official windows build

* Harden schedule test for slow windows timers

This test seems to be a bit flaky on windows, so give it more time to converge

For simplicity, perform parallelization of embedding requests in the API handler instead of offloading this to the subprocess runner. This keeps the scheduling story simpler as it builds on existing parallel requests, similar to existing text completion functionality.

* add prompt tokens to embed response

* rm slog

* metrics

* types

* prompt n

* clean up

* reset submodule

* update tests

* test name

* list metrics

In mult-brand GPU setups, if we couldn't fully load the model we
would fall through the scheduler and mistakenly try to load across
a mix of brands.  This makes sure we find the set of GPU(s) that
best fit for the partial load.

* Initial Batch Embedding

* Revert "Initial Batch Embedding"

This reverts commit c22d54895a280b54c727279d85a5fc94defb5a29.

* Initial Draft

* mock up notes

* api/embed draft

* add server function

* check normalization

* clean up

* normalization

* playing around with truncate stuff

* Truncation

* Truncation

* move normalization to go

* Integration Test Template

* Truncation Integration Tests

* Clean up

* use float32

* move normalize

* move normalize test

* refactoring

* integration float32

* input handling and handler testing

* Refactoring of legacy and new

* clear comments

* merge conflicts

* touches

* embedding type 64

* merge conflicts

* fix hanging on single string

* refactoring

* test values

* set context length

* clean up

* testing clean up

* testing clean up

* remove function closure

* Revert "remove function closure"

This reverts commit 55d48c6ed17abe42e7a122e69d603ef0c1506787.

* remove function closure

* remove redundant error check

* clean up

* more clean up

* clean up

This breaks up some of the test scenarios to create a
more reliable set of tests, as well as adding a little more
coverage.

This change fixes the handling of keep_alive so that if client
request omits the setting, we only set this on initial load.  Once
the model is loaded, if new requests leave this unset, we'll keep
whatever keep_alive was there.

Users may not realize the siny new model they're trying to load
fits on their disk, but can't load into system+GPU memory.  Today
we crash, but with this fix, we'll give them a better error message
before even trying to load it.

Previously, some costly things were causing the loading of GGUF files
and their metadata and tensor information to be VERY slow:

  * Too many allocations when decoding strings
  * Hitting disk for each read of each key and value, resulting in a
    not-okay amount of syscalls/disk I/O.

The show API is now down to 33ms from 800ms+ for llama3 on a macbook pro
m3.

This commit also prevents collecting large arrays of values when
decoding GGUFs (if desired). When such keys are encountered, their
values are null, and are encoded as such in JSON.

Also, this fixes a broken test that was not encoding valid GGUF.

This adjusts our default settings to enable multiple models and parallel
requests to a single model.  Users can still override these by the same
env var settings as before.  Parallel has a direct impact on
num_ctx, which in turn can have a significant impact on small VRAM GPUs
so this change also refines the algorithm so that when parallel is not
explicitly set by the user, we try to find a reasonable default that fits
the model on their GPU(s).  As before, multiple models will only load
concurrently if they fully fit in VRAM.

Still not complete, needs some refinement to our prediction to understand the
discrete GPUs available space so we can see how many layers fit in each one
since we can't split one layer across multiple GPUs we can't treat free space
as one logical block

* put flash attention behind flag for now

* add test

* remove print

* up timeout for sheduler tests