# What does this PR do?
the LiteLLMOpenAIMixin provides support for reading key from provider
data (headers users send).
this adds the same functionality to the OpenAIMixin.
this is infrastructure for migrating providers.
## Test Plan
ci w/ new tests
# What does this PR do?
Refs: https://github.com/llamastack/llama-stack/issues/3420
When telemetry is enabled the router uncondionally expects the usage
attribute to be availble and fails if it is not present.
Usage is not currently being requested by litellm_openai_mixin.py for
streaming requests when using the responses API which means that
providers like vertexai fail if telemetry is enabled and streaming is
used.
This is part of the required fix. Other part is in liteLLM, will plan to
submit PR for that soon.
## Test Plan
I applied this change along with the change for litellm in a llama stack
deployment and validated that I could make streaming requests through
the responses API to a gemini model and they would succeed instead of
failing due to the missing usage attribute when telemetry is enabled.
Signed-off-by: Michael Dawson <midawson@redhat.com>
# What does this PR do?
<!-- Provide a short summary of what this PR does and why. Link to
relevant issues if applicable. -->
Add items and title to ToolParameter/ToolParamDefinition. Adding items
will resolve the issue that occurs with Gemini LLM when an MCP tool has
array-type properties.
<!-- If resolving an issue, uncomment and update the line below -->
<!-- Closes #[issue-number] -->
## Test Plan
<!-- Describe the tests you ran to verify your changes with result
summaries. *Provide clear instructions so the plan can be easily
re-executed.* -->
Unite test cases will be added.
---------
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>
Co-authored-by: Kai Wu <kaiwu@meta.com>
Co-authored-by: Ashwin Bharambe <ashwin.bharambe@gmail.com>
# What does this PR do?
unpublish (make unavailable to users) the following apis -
- `/v1/inference/completion`, replaced by `/v1/openai/v1/completions`
- `/v1/inference/chat-completion`, replaced by
`/v1/openai/v1/chat/completions`
- `/v1/inference/embeddings`, replaced by `/v1/openai/v1/embeddings`
- `/v1/inference/batch-completion`, replaced by `/v1/openai/v1/batches`
- `/v1/inference/batch-chat-completion`, replaced by
`/v1/openai/v1/batches`
note: the implementations are still available for internal use, e.g.
agents uses chat-completion.
# What does this PR do?
simplify Ollama inference adapter by -
- moving image_url download code to OpenAIMixin
- being a ModelRegistryHelper instead of having one (mypy blocks
check_model_availability method assignment)
## Test Plan
- add unit tests for new download feature
- add integration tests for openai_chat_completion w/ image_url (close
test gap)
# What does this PR do?
- remove auto-download of ollama embedding models
- add embedding model metadata to dynamic listing w/ unit test
- add support and tests for allowed_models
- removed inference provider models.py files where dynamic listing is
enabled
- store embedding metadata in embedding_model_metadata field on
inference providers
- make model_entries optional on ModelRegistryHelper and
LiteLLMOpenAIMixin
- make OpenAIMixin a ModelRegistryHelper
- skip base64 embedding test for remote::ollama, always returns floats
- only use OpenAI client for ollama model listing
- remove unused build_model_entry function
- remove unused get_huggingface_repo function
## Test Plan
ci w/ new tests
# What does this PR do?
the openai_embeddings method on OpenAIMixin was returning the provider's
model id instead of the llama stack name
## Test Plan
before -
```
$ ./scripts/integration-tests.sh --stack-config server:ci-tests --setup gpt --subdirs inference --inference-mode live --pattern test_openai_embeddings_single_string
...
FAILED tests/integration/inference/test_openai_embeddings.py::test_openai_embeddings_single_string[openai_client-emb=openai/text-embedding-3-small] - AssertionError: assert 'text-embedding-3-small' == 'openai/text-...dding-3-small'
FAILED tests/integration/inference/test_openai_embeddings.py::test_openai_embeddings_single_string[llama_stack_client-emb=openai/text-embedding-3-small] - AssertionError: assert 'text-embedding-3-small' == 'openai/text-...dding-3-small'
========================================== 2 failed, 95 deselected, 4 warnings in 3.87s ===========================================
```
after -
```
$ ./scripts/integration-tests.sh --stack-config server:ci-tests --setup gpt --subdirs inference --inference-mode live --pattern test_openai_embeddings_single_string ...
========================================== 2 passed, 95 deselected, 4 warnings in 2.12s ===========================================
```
# What does this PR do?
<!-- Provide a short summary of what this PR does and why. Link to
relevant issues if applicable. -->
change ModelRegistryHelper to use ProviderModelEntry instead of
hardcoded ModelType.llm which fixed issue #3330.
<!-- If resolving an issue, uncomment and update the line below -->
<!-- Closes #[3330] -->
## Test Plan
<!-- Describe the tests you ran to verify your changes with result
summaries. *Provide clear instructions so the plan can be easily
re-executed.* -->
1. open llama-stack server
```
uv sync --python 3.12
source .venv/bin/activate
uv run llama stack build --distro starter --image-type venv --run
```
2.Used following script to test
```
from llama_stack_client import LlamaStackClient
import os
def test_openai_embedding_type():
client = LlamaStackClient(
base_url=os.environ.get("LLAMA_STACK_ENDPOINT", "http://localhost:8321"),
provider_data={
"openai_api_key": os.environ.get("OPENAI_API_KEY", ""),
},
)
model = client.models.retrieve("openai/text-embedding-3-small")
print(model)
assert model.identifier == "openai/text-embedding-3-small"
assert model.model_type == "embedding"
test_openai_embedding_type()
```
logs:
```
python test_openai.py
INFO:httpx:HTTP Request: GET http://localhost:8321/v1/models/openai/text-embedding-3-small "HTTP/1.1 200 OK"
Model(identifier='openai/text-embedding-3-small', metadata={'embedding_dimension': 1536.0, 'context_length': 8192.0}, api_model_type='embedding', provider_id='openai', type='model', provider_resource_id='text-embedding-3-small', owner=None, source='listed_from_provider', model_type='embedding')
```
# What does this PR do?
This PR is generated with AI and reviewed by me.
Refactors the AuthorizedSqlStore class to store the access policy as an
instance variable rather than passing it as a parameter to each method
call. This simplifies the API.
# Test Plan
existing tests
# What does this PR do?
this replaces the static model listing for any provider using
OpenAIMixin
currently -
- anthropic
- azure openai
- gemini
- groq
- llama-api
- nvidia
- openai
- sambanova
- tgi
- vertexai
- vllm
- not changed: together has its own impl
## Test Plan
- new unit tests
- manual for llama-api, openai, groq, gemini
```
for provider in llama-openai-compat openai groq gemini; do
uv run llama stack build --image-type venv --providers inference=remote::provider --run &
uv run --with llama-stack-client llama-stack-client models list | grep Total
```
results (17 sep 2025):
- llama-api: 4
- openai: 86
- groq: 21
- gemini: 66
closes#3467
# What does this PR do?
adds dynamic model support to TGI
add new overwrite_completion_id feature to OpenAIMixin to deal with TGI
always returning id=""
## Test Plan
tgi: `docker run --gpus all --shm-size 1g -p 8080:80 -v /data:/data
ghcr.io/huggingface/text-generation-inference --model-id
Qwen/Qwen3-0.6B`
stack: `TGI_URL=http://localhost:8080 uv run llama stack build
--image-type venv --distro ci-tests --run`
test: `./scripts/integration-tests.sh --stack-config
http://localhost:8321 --setup tgi --subdirs inference --pattern openai`
# What does this PR do?
update vLLM inference provider to use OpenAIMixin for openai-compat
functions
inference recordings from Qwen3-0.6B and vLLM 0.8.3 -
```
docker run --gpus all -v ~/.cache/huggingface:/root/.cache/huggingface -p 8000:8000 --ipc=host \
vllm/vllm-openai:latest \
--model Qwen/Qwen3-0.6B --enable-auto-tool-choice --tool-call-parser hermes
```
## Test Plan
```
./scripts/integration-tests.sh --stack-config server:ci-tests --setup vllm --subdirs inference
```
# What does this PR do?
Duplicate chat completion IDs can be generated during tests especially
if they are replaying recorded responses across different tests. No need
to warn or error under those circumstances. In the wild, this is not
likely to happen at all (no evidence) so we aren't really hiding any
problem.
# What does this PR do?
Adds a write worker queue for writes to inference store. This avoids
overwhelming request processing with slow inference writes.
## Test Plan
Benchmark:
```
cd /docs/source/distributions/k8s-benchmark
# start mock server
python openai-mock-server.py --port 8000
# start stack server
LLAMA_STACK_LOGGING="all=WARNING" uv run --with llama-stack python -m llama_stack.core.server.server docs/source/distributions/k8s-benchmark/stack_run_config.yaml
# run benchmark script
uv run python3 benchmark.py --duration 120 --concurrent 50 --base-url=http://localhost:8321/v1/openai/v1 --model=vllm-inference/meta-llama/Llama-3.2-3B-Instruct
```
## RPS from 21 -> 57
- Wrap model loading with asyncio.to_thread() to prevent blocking during
model download/initialization
- Wrap encoding operations with asyncio.to_thread() to run in background
thread
- Convert _load_sentence_transformer_model() to async method
This ensures the async event loop remains responsive during embedding
operations.
Closes: #3332
Signed-off-by: Derek Higgins <derekh@redhat.com>
Co-authored-by: Francisco Arceo <arceofrancisco@gmail.com>
# What does this PR do?
closes https://github.com/llamastack/llama-stack/issues/3236
mypy considered our default implementations (raise NotImplementedError)
to be trivial. the result was we implemented the same stubs in
providers.
this change puts enough into the default impls so mypy considers them
non-trivial. this allows us to remove the duplicate implementations.
# What does this PR do?
<!-- Provide a short summary of what this PR does and why. Link to
relevant issues if applicable. -->
This PR renames categories of llama_stack loggers.
This PR aligns logging categories as per the package name, as well as
reviews from initial
https://github.com/meta-llama/llama-stack/pull/2868. This is a follow up
to #3061.
<!-- If resolving an issue, uncomment and update the line below -->
<!-- Closes #[issue-number] -->
Replaces https://github.com/meta-llama/llama-stack/pull/2868
Part of https://github.com/meta-llama/llama-stack/issues/2865
cc @leseb @rhuss
Signed-off-by: Mustafa Elbehery <melbeher@redhat.com>
# What does this PR do?
This PR adds a step in pre-commit to enforce using `llama_stack` logger.
Currently, various parts of the code base uses different loggers. As a
custom `llama_stack` logger exist and used in the codebase, it is better
to standardize its utilization.
Signed-off-by: Mustafa Elbehery <melbeher@redhat.com>
Co-authored-by: Matthew Farrellee <matt@cs.wisc.edu>
This OpenAI client release
0843a11164
ends up breaking litellm
169a17400f/litellm/types/llms/openai.py (L40)
Update the dependency pin. Also make the imports a bit more defensive
anyhow if something else during `llama stack build` ends up moving
openai to a previous version.
## Test Plan
Run pre-release script integration tests.
The OpenAI compatibility layer was incorrectly importing
ChatCompletionMessageToolCallParam instead of the
ChatCompletionMessageFunctionToolCall class. This caused "Cannot
instantiate typing.Union" errors when processing agent requests with
tool calls.
Closes: #3141
Signed-off-by: Derek Higgins <derekh@redhat.com>
# What does this PR do?
I found a few issues while adding new metrics for various APIs:
currently metrics are only propagated in `chat_completion` and
`completion`
since most providers use the `openai_..` routes as the default in
`llama-stack-client inference chat-completion`, metrics are currently
not working as expected.
in order to get them working the following had to be done:
1. get the completion as usual
2. use new `openai_` versions of the metric gathering functions which
use `.usage` from the `OpenAI..` response types to gather the metrics
which are already populated.
3. define a `stream_generator` which counts the tokens and computes the
metrics (only for stream=True)
5. add metrics to response
NOTE: I could not add metrics to `openai_completion` where stream=True
because that ONLY returns an `OpenAICompletion` not an AsyncGenerator
that we can manipulate.
acquire the lock, and add event to the span as the other `_log_...`
methods do
some new output:
`llama-stack-client inference chat-completion --message hi`
<img width="2416" height="425" alt="Screenshot 2025-07-16 at 8 28 20 AM"
src="https://github.com/user-attachments/assets/ccdf1643-a184-4ddd-9641-d426c4d51326"
/>
and in the client:
<img width="763" height="319" alt="Screenshot 2025-07-16 at 8 28 32 AM"
src="https://github.com/user-attachments/assets/6bceb811-5201-47e9-9e16-8130f0d60007"
/>
these were not previously being recorded nor were they being printed to
the server due to the improper console sink handling
---------
Signed-off-by: Charlie Doern <cdoern@redhat.com>
# What does this PR do?
closes#2995
update SambaNovaInferenceAdapter to efficiently use LiteLLMOpenAIMixin
## Test Plan
```
$ uv run pytest -s -v tests/integration/inference --stack-config inference=sambanova --text-model sambanova/Meta-Llama-3.1-8B-Instruct
...
======================== 10 passed, 84 skipped, 3 xfailed, 51 warnings in 8.14s ========================
```
# What does this PR do?
Improve user experience by providing specific guidance when no API key
is available, showing both provider data header and config options with
the correct field name for each provider.
Also adds comprehensive test coverage for API key resolution scenarios.
addresses #2990 for providers using litellm openai mixin
## Test Plan
`./scripts/unit-tests.sh
tests/unit/providers/inference/test_litellm_openai_mixin.py`
# What does this PR do?
This enhancement allows inference providers using LiteLLMOpenAIMixin to
validate model availability against LiteLLM's official provider model
listings, improving reliability and user experience when working with
different AI service providers.
- Add litellm_provider_name parameter to LiteLLMOpenAIMixin constructor
- Add check_model_availability method to LiteLLMOpenAIMixin using
litellm.models_by_provider
- Update Gemini, Groq, and SambaNova inference adapters to pass
litellm_provider_name
## Test Plan
standard CI.
This flips #2823 and #2805 by making the Stack periodically query the
providers for models rather than the providers going behind the back and
calling "register" on to the registry themselves. This also adds support
for model listing for all other providers via `ModelRegistryHelper`.
Once this is done, we do not need to manually list or register models
via `run.yaml` and it will remove both noise and annoyance (setting
`INFERENCE_MODEL` environment variables, for example) from the new user
experience.
In addition, it adds a configuration variable `allowed_models` which can
be used to optionally restrict the set of models exposed from a
provider.
# What does this PR do?
add an `OpenAIMixin` for use by inference providers who remote endpoints
support an OpenAI compatible API.
use is demonstrated by refactoring
- OpenAIInferenceAdapter
- NVIDIAInferenceAdapter (adds embedding support)
- LlamaCompatInferenceAdapter
## Test Plan
existing unit and integration tests
This PR updates model registration and lookup behavior to be slightly
more general / flexible. See
https://github.com/meta-llama/llama-stack/issues/2843 for more details.
Note that this change is backwards compatible given the design of the
`lookup_model()` method.
## Test Plan
Added unit tests
# What does this PR do?
inference providers each have a static list of supported / known models.
some also have access to a dynamic list of currently available models.
this change gives prodivers using the ModelRegistryHelper the ability to
combine their static and dynamic lists.
for instance, OpenAIInferenceAdapter can implement
```
def query_available_models(self) -> list[str]:
return [entry.model for entry in self.openai_client.models.list()]
```
to augment its static list w/ a current list from openai.
## Test Plan
scripts/unit-test.sh
# What does this PR do?
We are now testing the safety capability with the starter image. This
includes a few changes:
* Enable the safety integration test
* Relax the shield model requirements from llama-guard to make it work
with llama-guard3:8b coming from Ollama
* Expose a shield for each inference provider in the starter distro. The
shield will only be registered if the provider is enabled.
Closes: https://github.com/meta-llama/llama-stack/issues/2528
Signed-off-by: Sébastien Han <seb@redhat.com>
## What does this PR do?
Ollama does not support remote images. Only local file paths OR base64
inputs are supported. This PR ensures that the Stack downloads remote
images and passes the base64 down to the inference engine.
## Test Plan
Added a test cases for Responses and ran it for both `fireworks` and
`ollama` providers.
# What does this PR do?
- llama_stack/exceptions.py: Add UnsupportedModelError class
- remote inference ollama.py and utils/inference/model_registry.py:
Changed ValueError in favor of UnsupportedModelError
- utils/inference/litellm_openai_mixin.py: remove `register_model`
function implementation from `LiteLLMOpenAIMixin` class. Now uses the
parent class `ModelRegistryHelper`'s function implementation
Closes#2517
## Test Plan
1. Create a new `test_run_openai.yaml` and paste the following config in
it:
```yaml
version: '2'
image_name: test-image
apis:
- inference
providers:
inference:
- provider_id: openai
provider_type: remote::openai
config:
max_tokens: 8192
models:
- metadata: {}
model_id: "non-existent-model"
provider_id: openai
model_type: llm
server:
port: 8321
```
And run the server with:
```bash
uv run llama stack run test_run_openai.yaml
```
You should now get a `llama_stack.exceptions.UnsupportedModelError` with
the supported list of models in the error message.
---
Tested for the following remote inference providers, and they all raise
the `UnsupportedModelError`:
- Anthropic
- Cerebras
- Fireworks
- Gemini
- Groq
- Ollama
- OpenAI
- SambaNova
- Together
- Watsonx
---------
Co-authored-by: Rohan Awhad <rawhad@redhat.com>
# What does this PR do?
* Given that our API packages use "import *" in `__init.py__` we don't
need to do `from llama_stack.apis.models.models` but simply from
llama_stack.apis.models. The decision to use `import *` is debatable and
should probably be revisited at one point.
* Remove unneeded Ruff F401 rule
* Consolidate Ruff F403 rule in the pyprojectfrom
llama_stack.apis.models.models
Signed-off-by: Sébastien Han <seb@redhat.com>
# What does this PR do?
closes#2522
## Test Plan
added integration test
LLAMA_STACK_CONFIG=http://localhost:8321 pytest -v
tests/integration/agents/test_openai_responses.py --text-model
"accounts/fireworks/models/llama-v3p3-70b-instruct" -vv -k
'function_call'
# What does this PR do?
Inference/Response stores now store user attributes when inserting, and
respects them when fetching.
## Test Plan
pytest tests/unit/utils/test_sqlstore.py
# What does this PR do?
dropped python3.10, updated pyproject and dependencies, and also removed
some blocks of code with special handling for enum.StrEnum
Closes#2458
Signed-off-by: Charlie Doern <cdoern@redhat.com>
For code completion apps need "fill in the middle" capabilities.
Added option of `suffix` to `openai_completion` to enable this.
Updated ollama provider to showcase the same.
### Test Plan
```
pytest -sv --stack-config="inference=ollama" tests/integration/inference/test_openai_completion.py --text-model qwen2.5-coder:1.5b -k test_openai_completion_non_streaming_suffix
```
### OpenAI Sample script
```
from openai import OpenAI
client = OpenAI(base_url="http://localhost:8321/v1/openai/v1")
response = client.completions.create(
model="qwen2.5-coder:1.5b",
prompt="The capital of ",
suffix="is Paris.",
max_tokens=10,
)
print(response.choices[0].text)
```
### Output
```
France is ____.
To answer this question, we
```
# What does this PR do?
This PR adds OpenAI compatibility for Ollama embeddings. Closes
https://github.com/meta-llama/llama-stack/issues/2428
Summary of changes:
- `llama_stack/providers/remote/inference/ollama/ollama.py`
- Implements the OpenAI embeddings endpoint for Ollama, replacing the
NotImplementedError with a full function that validates the model,
prepares parameters, calls the client, encodes embedding data
(optionally in base64), and returns a correctly structured response.
- Updates import statements to include the new embedding response
utilities.
- `llama_stack/providers/utils/inference/litellm_openai_mixin.py`
- Refactors the embedding data encoding logic to use a new shared
utility (`b64_encode_openai_embeddings_response`) instead of inline
base64 encoding and packing logic.
- Cleans up imports accordingly.
- `llama_stack/providers/utils/inference/openai_compat.py`
- Adds `b64_encode_openai_embeddings_response` to handle encoding OpenAI
embedding outputs (including base64 support) in a reusable way.
- Adds `prepare_openai_embeddings_params` utility for standardizing
embedding parameter preparation.
- Updates imports to include the new embedding data class.
- `tests/integration/inference/test_openai_embeddings.py`
- Removes `"remote::ollama"` from the list of providers that skip OpenAI
embeddings tests, since support is now implemented.
## Note
There was one minor issue, which required me to override the
`OpenAIEmbeddingsResponse.model` name with
`self._get_model(model).identifier` name, which is very unsatisfying.
## Test Plan
Unit Tests and integration tests
---------
Signed-off-by: Francisco Javier Arceo <farceo@redhat.com>
# What does this PR do?
Adds a new endpoint that is compatible with OpenAI for embeddings api.
`/openai/v1/embeddings`
Added providers for OpenAI, LiteLLM and SentenceTransformer.
## Test Plan
```
LLAMA_STACK_CONFIG=http://localhost:8321 pytest -sv tests/integration/inference/test_openai_embeddings.py --embedding-model all-MiniLM-L6-v2,text-embedding-3-small,gemini/text-embedding-004
```
