GitLab

This change adds a test to catch a regression in schema_to_grammar where
the order of keys in the JSON schema is not preserved in the generated
grammar, which is critical for step-by-step reasoning.

Sometimes the KV cache requires defragmentation even without
triggering the threshold heuristic. In this case, decoding
will not being able to find a KV cache slot. This is particularly
difficult for the caller to handle if it happens in between
ubatches. To avoid this, we should immediately trigger a defrag.

In addition, a heavily fragmented cache can require more than
max_moves to defragment. Currently, we stop when we hit the limit
but this can leave a cache that still does not have adequate space
even after defragmentation is triggered. Instead, we should do
multiple batches of processing until everything is complete.

Fixes #7949

This fixes another regression in the previous commit that fixed other
known bugs.

Changes in #8002 introduced fixes for bugs with mangling JSON Schemas.
It also fixed a bug where the server would silently fail when clients
requested invalid formats. It also, unfortunately, introduced a bug
where the server would reject requests with an empty format, which
should be allowed.

The change in #8127 updated the code to allow the empty format, but also
reintroduced the regression where the server would silently fail when
the format was set, but invalid.

This commit fixes both regressions. The server does not reject the empty
format, but it does reject invalid formats. It also adds tests to help
us catch regressions in the future.

Also, the updated code provides a more detailed error message when a
client sends a non-empty, but invalid format, echoing the invalid format
in the response.

This commits also takes the opportunity to remove superfluous linter
checks.

In 0.5.2 we simplified packaging to have avx only for macos x86.  It looks like
there may still be some non-AVX systems out there, so this puts back the prior
logic of building no-AVX for the primary binary, and now 2 runners for avx and avx2.
These will be packaged in the App bundle only, so the stand-alone binary will now be
without AVX support on macos.  On arm, we'll also see these runners reported
as available in the log, but they're dormant and will never be used at runtime.

readme: example/get started guide for pgai with Ollama

* docs: switch around database integrations order and link to quickstart

* docs: link to blog post in example readme

* chore: link to main readme

* readme: removing example to link externally

readme: removing example to link externally so we don't have to keep this example up-to-date

---------

Refactor mllama image processing code, and add pixtral and qwen2vl

This puts the low-level runner logging back on stderr for consistency with prior releases

* openai: return usage as final chunk for streams

---------
Co-authored-by: ParthSareen <parth.sareen@ollama.com>

Fixes #7944

Pass through the version override so the makefiles use it

Previously we decoded and re-encoded JSON schemas during validation,
which served no purpose since json.RawMessage already validates JSON
syntax. Worse, the re-encoding lost field ordering from the original
schema, which affects inference quality during step-by-step reasoning.

While fixing this ordering issue by using json.RawMessage directly,
testing revealed that schema_to_grammar (from llama.cpp) also fails to
preserve field order during grammar generation. This appears to be the
root cause of inference degradation.

This change prevents us from mangling the user's original schema order,
but we still need to address the ordering issue in schema_to_grammar.
That will be a separate change.

Updates #7978

upload-artifacts strips off leading common paths so when
the ./build/ artifacts were removed, the ./dist/windows-amd64
prefix became common and was stripped, making the
later download-artifacts place them in the wrong location

The new build embeds the arm runner in the
main binary, so there is no longer a lib/ollama

Remove no longer relevant build log dir

Signed-off-by: zhanluxianshen <zhanluxianshen@163.com>

The final implementation of #7499 removed dynamic vector requirements
in favor of a simpler filename based model, and this was left over logic that
is no longer needed.

The "F" was missing.

* llama: wire up builtin runner

This adds a new entrypoint into the ollama CLI to run the cgo built runner.
On Mac arm64, this will have GPU support, but on all other platforms it will
be the lowest common denominator CPU build.  After we fully transition
to the new Go runners more tech-debt can be removed and we can stop building
the "default" runner via make and rely on the builtin always.

* build: Make target improvements

Add a few new targets and help for building locally.
This also adjusts the runner lookup to favor local builds, then
runners relative to the executable, and finally payloads.

* Support customized CPU flags for runners

This implements a simplified custom CPU flags pattern for the runners.
When built without overrides, the runner name contains the vector flag
we check for (AVX) to ensure we don't try to run on unsupported systems
and crash.  If the user builds a customized set, we omit the naming
scheme and don't check for compatibility.  This avoids checking
requirements at runtime, so that logic has been removed as well.  This
can be used to build GPU runners with no vector flags, or CPU/GPU
runners with additional flags (e.g. AVX512) enabled.

* Use relative paths

If the user checks out the repo in a path that contains spaces, make gets
really confused so use relative paths for everything in-repo to avoid breakage.

* Remove payloads from main binary

* install: clean up prior libraries

This removes support for v0.3.6 and older versions (before the tar bundle)
and ensures we clean up prior libraries before extracting the bundle(s).
Without this change, runners and dependent libraries could leak when we
update and lead to subtle runtime errors.