# What does this PR do?
The telemetry provider configs is the only one who leverages the env var
`SQLITE_DB_PATH` for pointing to persistent data in the respective
templates, whereas usually `SQLITE_STORE_DIR` is used.
This PR modifies the `sqlite_db_path` in various telemetry configuration
files to use the environment variable `SQLITE_STORE_DIR` instead of
`SQLITE_DB_PATH`. This change ensures that _only_ the SQLITE_STORE_DIR
needs to be set to point to a different persistence location for
providers.
All references to `SQLITE_DB_PATH` have been removed.
Another improvement could be to move `sqlite_db_path` to `db_path` in
the telemetry provider config, to align with the other provider
configurations. That could be done by another PR (if wanted).
# What does this PR do?
This provides an initial [OpenAI Responses
API](https://platform.openai.com/docs/api-reference/responses)
implementation. The API is not yet complete, and this is more a
proof-of-concept to show how we can store responses in our key-value
stores and use them to support the Responses API concepts like
`previous_response_id`.
## Test Plan
I've added a new
`tests/integration/openai_responses/test_openai_responses.py` as part of
a test-driven development for this new API. I'm only testing this
locally with the remote-vllm provider for now, but it should work with
any of our inference providers since the only API it requires out of the
inference provider is the `openai_chat_completion` endpoint.
```
VLLM_URL="http://localhost:8000/v1" \
INFERENCE_MODEL="meta-llama/Llama-3.2-3B-Instruct" \
llama stack build --template remote-vllm --image-type venv --run
```
```
LLAMA_STACK_CONFIG="http://localhost:8321" \
python -m pytest -v \
tests/integration/openai_responses/test_openai_responses.py \
--text-model "meta-llama/Llama-3.2-3B-Instruct"
```
---------
Signed-off-by: Ben Browning <bbrownin@redhat.com>
Co-authored-by: Ashwin Bharambe <ashwin.bharambe@gmail.com>
# What does this PR do?
This commit adds a new authentication system to the Llama Stack server
with support for Kubernetes and custom authentication providers. Key
changes include:
- Implemented KubernetesAuthProvider for validating Kubernetes service
account tokens
- Implemented CustomAuthProvider for validating tokens against external
endpoints - this is the same code that was already present.
- Added test for Kubernetes
- Updated server configuration to support authentication settings
- Added documentation for authentication configuration and usage
The authentication system supports:
- Bearer token validation
- Kubernetes service account token validation
- Custom authentication endpoints
## Test Plan
Setup a Kube cluster using Kind or Minikube.
Run a server with:
```
server:
port: 8321
auth:
provider_type: kubernetes
config:
api_server_url: http://url
ca_cert_path: path/to/cert (optional)
```
Run:
```
curl -s -L -H "Authorization: Bearer $(kubectl create token my-user)" http://127.0.0.1:8321/v1/providers
```
Or replace "my-user" with your service account.
Signed-off-by: Sébastien Han <seb@redhat.com>
# What does this PR do?
Implemetation of NeMO Datastore register, unregister API.
Open Issues:
- provider_id gets set to `localfs` in client.datasets.register() as it
is specified in routing_tables.py: DatasetsRoutingTable
see: #1860
Currently I have passed `"provider_id":"nvidia"` in metadata and have
parsed that in `DatasetsRoutingTable`
(Not the best approach, but just a quick workaround to make it work for
now.)
## Test Plan
- Unit test cases: `pytest
tests/unit/providers/nvidia/test_datastore.py`
```bash
========================================================== test session starts ===========================================================
platform linux -- Python 3.10.0, pytest-8.3.5, pluggy-1.5.0
rootdir: /home/ubuntu/llama-stack
configfile: pyproject.toml
plugins: anyio-4.9.0, asyncio-0.26.0, nbval-0.11.0, metadata-3.1.1, html-4.1.1, cov-6.1.0
asyncio: mode=strict, asyncio_default_fixture_loop_scope=None, asyncio_default_test_loop_scope=function
collected 2 items
tests/unit/providers/nvidia/test_datastore.py .. [100%]
============================================================ warnings summary ============================================================
====================================================== 2 passed, 1 warning in 0.84s ======================================================
```
cc: @dglogo, @mattf, @yanxi0830
# What does this PR do?
This addresses 2 bugs I ran into when launching a fine-tuning job with
the NVIDIA Adapter:
1. Session handling in `_make_request` helper function returns an error.
```
INFO: 127.0.0.1:55831 - "POST /v1/post-training/supervised-fine-tune HTTP/1.1" 500 Internal Server Error
16:11:45.643 [END] /v1/post-training/supervised-fine-tune [StatusCode.OK] (270.44ms)
16:11:45.643 [ERROR] Error executing endpoint route='/v1/post-training/supervised-fine-tune' method='post'
Traceback (most recent call last):
File "/Users/jgulabrai/Projects/forks/llama-stack/llama_stack/distribution/server/server.py", line 201, in endpoint
return await maybe_await(value)
File "/Users/jgulabrai/Projects/forks/llama-stack/llama_stack/distribution/server/server.py", line 161, in maybe_await
return await value
File "/Users/jgulabrai/Projects/forks/llama-stack/llama_stack/providers/remote/post_training/nvidia/post_training.py", line 408, in supervised_fine_tune
response = await self._make_request(
File "/Users/jgulabrai/Projects/forks/llama-stack/llama_stack/providers/remote/post_training/nvidia/post_training.py", line 98, in _make_request
async with self.session.request(method, url, params=params, json=json, **kwargs) as response:
File "/Users/jgulabrai/Projects/forks/llama-stack/.venv/lib/python3.10/site-packages/aiohttp/client.py", line 1425, in __aenter__
self._resp: _RetType = await self._coro
File "/Users/jgulabrai/Projects/forks/llama-stack/.venv/lib/python3.10/site-packages/aiohttp/client.py", line 579, in _request
handle = tm.start()
File "/Users/jgulabrai/Projects/forks/llama-stack/.venv/lib/python3.10/site-packages/aiohttp/helpers.py", line 587, in start
return self._loop.call_at(when, self.__call__)
File "/Library/Frameworks/Python.framework/Versions/3.10/lib/python3.10/asyncio/base_events.py", line 724, in call_at
self._check_closed()
File "/Library/Frameworks/Python.framework/Versions/3.10/lib/python3.10/asyncio/base_events.py", line 510, in _check_closed
raise RuntimeError('Event loop is closed')
RuntimeError: Event loop is closed
```
Note: This only occurred when initializing the client like so:
```
client = LlamaStackClient(
base_url="http://0.0.0.0:8321"
)
response = client.post_training.supervised_fine_tune(...) # Returns error
```
I didn't run into this issue when using the library client:
```
client = LlamaStackAsLibraryClient("nvidia")
client.initialize()
response = client.post_training.supervised_fine_tune(...) # Works fine
```
2. The `algorithm_config` param in `supervised_fine_tune` is parsed as a
`dict` when run from unit tests, but a Pydantic model when invoked using
the Llama Stack client. So, the call fails outside of unit tests:
```
INFO: 127.0.0.1:54024 - "POST /v1/post-training/supervised-fine-tune HTTP/1.1" 500 Internal Server Error
21:14:02.315 [END] /v1/post-training/supervised-fine-tune [StatusCode.OK] (71.18ms)
21:14:02.314 [ERROR] Error executing endpoint route='/v1/post-training/supervised-fine-tune' method='post'
Traceback (most recent call last):
File "/Users/jgulabrai/Projects/forks/llama-stack/llama_stack/distribution/server/server.py", line 205, in endpoint
return await maybe_await(value)
File "/Users/jgulabrai/Projects/forks/llama-stack/llama_stack/distribution/server/server.py", line 164, in maybe_await
return await value
File "/Users/jgulabrai/Projects/forks/llama-stack/llama_stack/providers/remote/post_training/nvidia/post_training.py", line 407, in supervised_fine_tune
"adapter_dim": algorithm_config.get("adapter_dim"),
File "/Users/jgulabrai/Projects/forks/llama-stack/.venv/lib/python3.10/site-packages/pydantic/main.py", line 891, in __getattr__
raise AttributeError(f'{type(self).__name__!r} object has no attribute {item!r}')
AttributeError: 'LoraFinetuningConfig' object has no attribute 'get'
```
The code assumes `algorithm_config` should be `dict`, so I just handle
both cases.
[//]: # (If resolving an issue, uncomment and update the line below)
[//]: # (Closes #[issue-number])
## Test Plan
1. I ran a local Llama Stack server with the necessary env vars:
```
lama stack run llama_stack/templates/nvidia/run.yaml --port 8321 --env ...
```
And invoked `supervised_fine_tune` to confirm neither of the errors
above occur.
```
client = LlamaStackClient(
base_url="http://0.0.0.0:8321"
)
response = client.post_training.supervised_fine_tune(...)
```
2. I confirmed the unit tests still pass: `./scripts/unit-tests.sh
tests/unit/providers/nvidia/test_supervised_fine_tuning.py`
[//]: # (## Documentation)
---------
Co-authored-by: Jash Gulabrai <jgulabrai@nvidia.com>
# What does this PR do?
Adds custom model registration functionality to NVIDIAInferenceAdapter
which let's the inference happen on:
- post-training model
- non-llama models in API Catalogue(behind
https://integrate.api.nvidia.com and endpoints compatible with
AyncOpenAI)
## Example Usage:
```python
from llama_stack.apis.models import Model, ModelType
from llama_stack.distribution.library_client import LlamaStackAsLibraryClient
client = LlamaStackAsLibraryClient("nvidia")
_ = client.initialize()
client.models.register(
model_id=model_name,
model_type=ModelType.llm,
provider_id="nvidia"
)
response = client.inference.chat_completion(
model_id=model_name,
messages=[{"role":"system","content":"You are a helpful assistant."},{"role":"user","content":"Write a limerick about the wonders of GPU computing."}],
)
```
## Test Plan
```bash
pytest tests/unit/providers/nvidia/test_supervised_fine_tuning.py
========================================================== test session starts ===========================================================
platform linux -- Python 3.10.0, pytest-8.3.5, pluggy-1.5.0
rootdir: /home/ubuntu/llama-stack
configfile: pyproject.toml
plugins: anyio-4.9.0
collected 6 items
tests/unit/providers/nvidia/test_supervised_fine_tuning.py ...... [100%]
============================================================ warnings summary ============================================================
../miniconda/envs/nvidia-1/lib/python3.10/site-packages/pydantic/fields.py:1076
/home/ubuntu/miniconda/envs/nvidia-1/lib/python3.10/site-packages/pydantic/fields.py:1076: PydanticDeprecatedSince20: Using extra keyword arguments on `Field` is deprecated and will be removed. Use `json_schema_extra` instead. (Extra keys: 'contentEncoding'). Deprecated in Pydantic V2.0 to be removed in V3.0. See Pydantic V2 Migration Guide at https://errors.pydantic.dev/2.11/migration/
warn(
-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
====================================================== 6 passed, 1 warning in 1.51s ======================================================
```
[//]: # (## Documentation)
Updated Readme.md
cc: @dglogo, @sumitb, @mattf
# What does this PR do?
This PR adds support for NVIDIA's NeMo Evaluator API to the Llama Stack
eval module. The integration enables users to evaluate models via the
Llama Stack interface.
## 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. Added unit tests and successfully ran from root of project:
`./scripts/unit-tests.sh tests/unit/providers/nvidia/test_eval.py`
```
tests/unit/providers/nvidia/test_eval.py::TestNVIDIAEvalImpl::test_job_cancel PASSED
tests/unit/providers/nvidia/test_eval.py::TestNVIDIAEvalImpl::test_job_result PASSED
tests/unit/providers/nvidia/test_eval.py::TestNVIDIAEvalImpl::test_job_status PASSED
tests/unit/providers/nvidia/test_eval.py::TestNVIDIAEvalImpl::test_register_benchmark PASSED
tests/unit/providers/nvidia/test_eval.py::TestNVIDIAEvalImpl::test_run_eval PASSED
```
2. Verified I could build the Llama Stack image: `LLAMA_STACK_DIR=$(pwd)
llama stack build --template nvidia --image-type venv`
Documentation added to
`llama_stack/providers/remote/eval/nvidia/README.md`
---------
Co-authored-by: Jash Gulabrai <jgulabrai@nvidia.com>
# What does this PR do?
This expands the `test_sse` test suite and fixes some edge cases with
bugs in our SSE error handling to ensure streaming clients always get a
proper error response.
First, we handle the case where a client disconnects before we actually
start streaming the response back. Previously we only handled the case
where a client disconnected as we were streaming the response, but there
was an edge case where a client disconnecting before we streamed any
response back did not trigger our logic to cleanly handle that
disconnect.
Second, we handle the case where an error is thrown from the server
before the actual async generator gets created from the provider. This
happens in scenarios like the newly merged OpenAI API input validation,
where we eagerly raise validation errors before returning the async
generator object that streams the responses back.
## Test Plan
Tested via:
```
python -m pytest -s -v tests/unit/server/test_sse.py
```
Both test cases failed before, and passed afterwards. The test cases
were written based on me experimenting with actual clients that would do
bad things like randomly disconnect or send invalid input in streaming
mode and I hit these two cases, where things were misbehaving in our
error handling.
Signed-off-by: Ben Browning <bbrownin@redhat.com>
Include the tool call details with the chat when doing Rag with Remote
vllm
Fixes: #1929
With this PR the tool call is included in the chat returned to vllm, the
model (meta-llama/Llama-3.1-8B-Instruct) the returns the answer as
expected.
Signed-off-by: Derek Higgins <derekh@redhat.com>
As part of the build process, we now include the generated run.yaml
(based of the provided build configuration file) into the container. We
updated the entrypoint to use this run configuration as well.
Given this simple distribution configuration:
```
# build.yaml
version: '2'
distribution_spec:
description: Use (an external) Ollama server for running LLM inference
providers:
inference:
- remote::ollama
vector_io:
- inline::faiss
safety:
- inline::llama-guard
agents:
- inline::meta-reference
telemetry:
- inline::meta-reference
eval:
- inline::meta-reference
datasetio:
- remote::huggingface
- inline::localfs
scoring:
- inline::basic
- inline::llm-as-judge
- inline::braintrust
tool_runtime:
- remote::brave-search
- remote::tavily-search
- inline::code-interpreter
- inline::rag-runtime
- remote::model-context-protocol
- remote::wolfram-alpha
container_image: "registry.access.redhat.com/ubi9"
image_type: container
image_name: test
```
Build it:
```
llama stack build --config build.yaml
```
Run it:
```
podman run --rm \
-p 8321:8321 \
-e OLLAMA_URL=http://host.containers.internal:11434 \
--name llama-stack-server \
localhost/leseb-test:0.2.2
```
Signed-off-by: Sébastien Han <seb@redhat.com>
# What does this PR do?
When clients called the Open AI API with invalid input that wasn't
caught by our own Pydantic API validation but instead only caught by the
backend inference provider, that backend inference provider was
returning a HTTP 400 error. However, we were wrapping that into a HTTP
500 error, obfuscating the actual issue from calling clients and
triggering OpenAI client retry logic.
This change adjusts our existing `translate_exception` method in
`server.py` to wrap `openai.BadRequestError` as HTTP 400 errors, passing
through the string representation of the error message to the calling
user so they can see the actual input validation error and correct it. I
tried changing this in a few other places, but ultimately
`translate_exception` was the only real place to handle this for both
streaming and non-streaming requests across all inference providers that
use the OpenAI server APIs.
This also tightens up our validation a bit for the OpenAI chat
completions API, to catch empty `messages` parameters, invalid
`tool_choice` parameters, invalid `tools` items, or passing
`tool_choice` when `tools` isn't given.
Lastly, this extends our OpenAI API chat completions verifications to
also check for consistent input validation across providers. Providers
behind Llama Stack should automatically pass all the new tests due to
the input validation added here, but some of the providers fail this
test when not run behind Llama Stack due to differences in how they
handle input validation and errors.
(Closes#1951)
## Test Plan
To test this, start an OpenAI API verification stack:
```
llama stack run --image-type venv tests/verifications/openai-api-verification-run.yaml
```
Then, run the new verification tests with your provider(s) of choice:
```
python -m pytest -s -v \
tests/verifications/openai_api/test_chat_completion.py \
--provider openai-llama-stack
python -m pytest -s -v \
tests/verifications/openai_api/test_chat_completion.py \
--provider together-llama-stack
```
Signed-off-by: Ben Browning <bbrownin@redhat.com>
# What does this PR do?
Previously, when a streaming client would disconnect before we were
finished streaming the entire response, an error like the below would
get raised from the `sse_generator` function in
`llama_stack/distribution/server/server.py`:
```
AttributeError: 'coroutine' object has no attribute 'aclose'. Did you mean: 'close'?
```
This was because we were calling `aclose` on a coroutine instead of the
awaited value from that coroutine. This change fixes that, so that we
save off the awaited value and then can call `aclose` on it if we
encounter an `asyncio.CancelledError`, like we see when a client
disconnects before we're finished streaming.
The other changes in here are to add a simple set of tests for the happy
path of our SSE streaming and this client disconnect path.
That unfortunately requires adding one more dependency into our unit
test section of pyproject.toml since `server.py` requires loading some
of the telemetry code for me to test this functionality.
## Test Plan
I wrote the tests in `tests/unit/server/test_sse.py` first, verified the
client disconnected test failed before my change, and that it passed
afterwards.
```
python -m pytest -s -v tests/unit/server/test_sse.py
```
Signed-off-by: Ben Browning <bbrownin@redhat.com>
Test plan:
python tests/verifications/generate_report.py --providers
fireworks,together,llama_meta_ref,openai
Co-authored-by: Eric Huang <erichuang@fb.com>
# What does this PR do?
Allow users to name an agent and use the name in telemetry instead of
relying on randomly generated agent_ids. This improves the developer
experience by making it easier to find specific agents in telemetry
logs.
Closes#1832
## Test Plan
- Added tests to verify the agent name is properly stored and retrieved
- Ran `uv run -- pytest -v
tests/integration/telemetry/test_telemetry.py::test_agent_name_filtering`
from the root of the project and made sure the tests pass
- Ran `uv run -- pytest -v
tests/integration/telemetry/test_telemetry.py::test_agent_query_spans`
to verify existing code without agent names still works correctly
## Use Example
```
agent = Agent(
llama_stack_client,
model=text_model_id,
name="CustomerSupportAgent", # New parameter
instructions="You are a helpful customer support assistant"
)
session_id = agent.create_session(f"test-session-{uuid4()}")
```
## Implementation Notes
- Agent names are optional string parameters with no additional
validation
- Names are not required to be unique - multiple agents can have the
same name
- The agent_id remains the unique identifier for an agent
---------
Co-authored-by: raghotham <raghotham@gmail.com>
# What does this PR do?
This PR handles the case where a Customization Job's status is
`unknown`. Since we don't map `unknown` to a valid `JobStatus`, the
PostTraining provider throws an exception when fetching/listing a job.
[//]: # (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.*]
`./scripts/unit-tests.sh
tests/unit/providers/nvidia/test_supervised_fine_tuning.py` succeeds
[//]: # (## Documentation)
Co-authored-by: Jash Gulabrai <jgulabrai@nvidia.com>
# What does this PR do?
Fixes a crash that occurred when building a stack as a container image
via the interactive wizard without supplying --template or --config.
- Root cause: template_or_config was None; only the container path
relies on that parameter, which later reaches subprocess.run() and
triggers
`TypeError: expected str, bytes or os.PathLike object, not NoneType.`
- Change: in `_run_stack_build_command_from_build_config` we now fall
back to the freshly‑written build‑spec file whenever both optional
sources are missing. Also adds a spy‑based unit test that asserts a
valid string path is passed to build_image() for container builds.
### Closes#1976
## Test Plan
- New unit test: test_build_path.py. Monkey‑patches build_image,
captures the fourth argument, and verifies it is a real path
- Manual smoke test:
```
llama stack build --image-type container
# answer wizard prompts
```
Build proceeds into Docker without raising the previous TypeError.
## Future Work
Harmonise `build_image` arguments so every image type receives the same
inputs, eliminating this asymmetric special‑case.
Fixes: #1955
Since 0.2.0, the vLLM gets an empty list (vs ``None``in 0.1.9 and
before) when there are no tools configured which causes the issue
described in #1955 p. This patch avoids sending the 'tools' param to the
vLLM altogether instead of an empty list.
It also adds a small unit test to avoid regressions.
The OpenAI
[specification](https://platform.openai.com/docs/api-reference/chat/create)
does not explicitly state that the list cannot be empty but I found this
out through experimentation and it might depend on the actual remote
vllm. In any case, as this parameter is Optional, is best to skip it
altogether if there's no tools configured.
Signed-off-by: Daniel Alvarez <dalvarez@redhat.com>
# What does this PR do?
## Test Plan
(myenv) ➜ llama-stack python tests/verifications/generate_report.py
--providers fireworks,together,openai --run-tests
f27f617629/tests/verifications/REPORT.md
# What does this PR do?
Now a separate thread is started to execute training jobs. Training
requests now return job ID before the job completes. (Which fixes API
timeouts for any jobs that take longer than a minute.)
Note: the scheduler code is meant to be spun out in the future into a
common provider service that can be reused for different APIs and
providers. It is also expected to back the /jobs API proposed here:
https://github.com/meta-llama/llama-stack/discussions/1238
Hence its somewhat generalized form which is expected to simplify its
adoption elsewhere in the future.
Note: this patch doesn't attempt to implement missing APIs (e.g. cancel
or job removal). This work will belong to follow-up PRs.
[//]: # (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.*]
Added unit tests for the scheduler module. For the API coverage, did
manual testing and was able to run a training cycle on GPU. The initial
call returned job ID before the training completed, as (now) expected.
Artifacts are returned as expected.
```
JobArtifactsResponse(checkpoints=[{'identifier': 'meta-llama/Llama-3.2-3B-Instruct-sft-0', 'created_at': '2025-03-07T22:45:19.892714', 'epoch': 0, 'post_training_job_id': 'test-job2ee77104-2fd3-4a4e-84cf-f83f8b8f1f50', 'path': '/home/ec2-user/.llama/checkpoints/meta-llama/Llama-3.2-3B-Instruct-sft-0', 'training_metrics': None}], job_uuid='test-job2ee77104-2fd3-4a4e-84cf-f83f8b8f1f50')
```
The integration test is currently disabled for the provider. I will look
into how it can be enabled in a different PR / issue context.
[//]: # (## Documentation)
Signed-off-by: Ihar Hrachyshka <ihar.hrachyshka@gmail.com>
# What does this PR do?
TLDR: Changes needed to get 100% passing tests for OpenAI API
verification tests when run against Llama Stack with the `together`,
`fireworks`, and `openai` providers. And `groq` is better than before,
at 88% passing.
This cleans up the OpenAI API support for image message types
(specifically `image_url` types) and handling of the `response_format`
chat completion parameter. Both of these required a few more Pydantic
model definitions in our Inference API, just to move from the
not-quite-right stubs I had in place to something fleshed out to match
the actual OpenAI API specs.
As part of testing this, I also found and fixed a bug in the litellm
implementation of openai_completion and openai_chat_completion, so the
providers based on those should actually be working now.
The method `prepare_openai_completion_params` in
`llama_stack/providers/utils/inference/openai_compat.py` was improved to
actually recursively clean up input parameters, including handling of
lists, dicts, and dumping of Pydantic models to dicts. These changes
were required to get to 100% passing tests on the OpenAI API
verification against the `openai` provider.
With the above, the together.ai provider was passing as well as it is
without Llama Stack. But, since we have Llama Stack in the middle, I
took the opportunity to clean up the together.ai provider so that it now
also passes the OpenAI API spec tests we have at 100%. That means
together.ai is now passing our verification test better when using an
OpenAI client talking to Llama Stack than it is when hitting together.ai
directly, without Llama Stack in the middle.
And, another round of work for Fireworks to improve translation of
incoming OpenAI chat completion requests to Llama Stack chat completion
requests gets the fireworks provider passing at 100%. The server-side
fireworks.ai tool calling support with OpenAI chat completions and Llama
4 models isn't great yet, but by pointing the OpenAI clients at Llama
Stack's API we can clean things up and get everything working as
expected for Llama 4 models.
## Test Plan
### OpenAI API Verification Tests
I ran the OpenAI API verification tests as below and 100% of the tests
passed.
First, start a Llama Stack server that runs the `openai` provider with
the `gpt-4o` and `gpt-4o-mini` models deployed. There's not a template
setup to do this out of the box, so I added a
`tests/verifications/openai-api-verification-run.yaml` to do this.
First, ensure you have the necessary API key environment variables set:
```
export TOGETHER_API_KEY="..."
export FIREWORKS_API_KEY="..."
export OPENAI_API_KEY="..."
```
Then, run a Llama Stack server that serves up all these providers:
```
llama stack run \
--image-type venv \
tests/verifications/openai-api-verification-run.yaml
```
Finally, generate a new verification report against all these providers,
both with and without the Llama Stack server in the middle.
```
python tests/verifications/generate_report.py \
--run-tests \
--provider \
together \
fireworks \
groq \
openai \
together-llama-stack \
fireworks-llama-stack \
groq-llama-stack \
openai-llama-stack
```
You'll see that most of the configurations with Llama Stack in the
middle now pass at 100%, even though some of them do not pass at 100%
when hitting the backend provider's API directly with an OpenAI client.
### OpenAI Completion Integration Tests with vLLM:
I also ran the smaller `test_openai_completion.py` test suite (that's
not yet merged with the verification tests) on multiple of the
providers, since I had to adjust the method signature of
openai_chat_completion a bit and thus had to touch lots of these
providers to match. Here's the tests I ran there, all passing:
```
VLLM_URL="http://localhost:8000/v1" INFERENCE_MODEL="meta-llama/Llama-3.2-3B-Instruct" llama stack build --template remote-vllm --image-type venv --run
```
in another terminal
```
LLAMA_STACK_CONFIG=http://localhost:8321 INFERENCE_MODEL="meta-llama/Llama-3.2-3B-Instruct" python -m pytest -v tests/integration/inference/test_openai_completion.py --text-model "meta-llama/Llama-3.2-3B-Instruct"
```
### OpenAI Completion Integration Tests with ollama
```
INFERENCE_MODEL="llama3.2:3b-instruct-q8_0" llama stack build --template ollama --image-type venv --run
```
in another terminal
```
LLAMA_STACK_CONFIG=http://localhost:8321 INFERENCE_MODEL="llama3.2:3b-instruct-q8_0" python -m pytest -v tests/integration/inference/test_openai_completion.py --text-model "llama3.2:3b-instruct-q8_0"
```
### OpenAI Completion Integration Tests with together.ai
```
INFERENCE_MODEL="meta-llama/Llama-3.2-3B-Instruct-Turbo" llama stack build --template together --image-type venv --run
```
in another terminal
```
LLAMA_STACK_CONFIG=http://localhost:8321 INFERENCE_MODEL="meta-llama/Llama-3.2-3B-Instruct-Turbo" python -m pytest -v tests/integration/inference/test_openai_completion.py --text-model "meta-llama/Llama-3.2-3B-Instruct-Turbo"
```
### OpenAI Completion Integration Tests with fireworks.ai
```
INFERENCE_MODEL="meta-llama/Llama-3.1-8B-Instruct" llama stack build --template fireworks --image-type venv --run
```
in another terminal
```
LLAMA_STACK_CONFIG=http://localhost:8321 INFERENCE_MODEL="meta-llama/Llama-3.1-8B-Instruct" python -m pytest -v tests/integration/inference/test_openai_completion.py --text-model "meta-llama/Llama-3.1-8B-Instruct"
---------
Signed-off-by: Ben Browning <bbrownin@redhat.com>
# What does this PR do?
This PR adds two methods to the Inference API:
- `batch_completion`
- `batch_chat_completion`
The motivation is for evaluations targeting a local inference engine
(like meta-reference or vllm) where batch APIs provide for a substantial
amount of acceleration.
Why did I not add this to `Api.batch_inference` though? That just
resulted in a _lot_ more book-keeping given the structure of Llama
Stack. Had I done that, I would have needed to create a notion of a
"batch model" resource, setup routing based on that, etc. This does not
sound ideal.
So what's the future of the batch inference API? I am not sure. Maybe we
can keep it for true _asynchronous_ execution. So you can submit
requests, and it can return a Job instance, etc.
## Test Plan
Run meta-reference-gpu using:
```bash
export INFERENCE_MODEL=meta-llama/Llama-4-Scout-17B-16E-Instruct
export INFERENCE_CHECKPOINT_DIR=../checkpoints/Llama-4-Scout-17B-16E-Instruct-20250331210000
export MODEL_PARALLEL_SIZE=4
export MAX_BATCH_SIZE=32
export MAX_SEQ_LEN=6144
LLAMA_MODELS_DEBUG=1 llama stack run meta-reference-gpu
```
Then run the batch inference test case.
# What does this PR do?
This stubs in some OpenAI server-side compatibility with three new
endpoints:
/v1/openai/v1/models
/v1/openai/v1/completions
/v1/openai/v1/chat/completions
This gives common inference apps using OpenAI clients the ability to
talk to Llama Stack using an endpoint like
http://localhost:8321/v1/openai/v1 .
The two "v1" instances in there isn't awesome, but the thinking is that
Llama Stack's API is v1 and then our OpenAI compatibility layer is
compatible with OpenAI V1. And, some OpenAI clients implicitly assume
the URL ends with "v1", so this gives maximum compatibility.
The openai models endpoint is implemented in the routing layer, and just
returns all the models Llama Stack knows about.
The following providers should be working with the new OpenAI
completions and chat/completions API:
* remote::anthropic (untested)
* remote::cerebras-openai-compat (untested)
* remote::fireworks (tested)
* remote::fireworks-openai-compat (untested)
* remote::gemini (untested)
* remote::groq-openai-compat (untested)
* remote::nvidia (tested)
* remote::ollama (tested)
* remote::openai (untested)
* remote::passthrough (untested)
* remote::sambanova-openai-compat (untested)
* remote::together (tested)
* remote::together-openai-compat (untested)
* remote::vllm (tested)
The goal to support this for every inference provider - proxying
directly to the provider's OpenAI endpoint for OpenAI-compatible
providers. For providers that don't have an OpenAI-compatible API, we'll
add a mixin to translate incoming OpenAI requests to Llama Stack
inference requests and translate the Llama Stack inference responses to
OpenAI responses.
This is related to #1817 but is a bit larger in scope than just chat
completions, as I have real use-cases that need the older completions
API as well.
## Test Plan
### vLLM
```
VLLM_URL="http://localhost:8000/v1" INFERENCE_MODEL="meta-llama/Llama-3.2-3B-Instruct" llama stack build --template remote-vllm --image-type venv --run
LLAMA_STACK_CONFIG=http://localhost:8321 INFERENCE_MODEL="meta-llama/Llama-3.2-3B-Instruct" python -m pytest -v tests/integration/inference/test_openai_completion.py --text-model "meta-llama/Llama-3.2-3B-Instruct"
```
### ollama
```
INFERENCE_MODEL="llama3.2:3b-instruct-q8_0" llama stack build --template ollama --image-type venv --run
LLAMA_STACK_CONFIG=http://localhost:8321 INFERENCE_MODEL="llama3.2:3b-instruct-q8_0" python -m pytest -v tests/integration/inference/test_openai_completion.py --text-model "llama3.2:3b-instruct-q8_0"
```
## Documentation
Run a Llama Stack distribution that uses one of the providers mentioned
in the list above. Then, use your favorite OpenAI client to send
completion or chat completion requests with the base_url set to
http://localhost:8321/v1/openai/v1 . Replace "localhost:8321" with the
host and port of your Llama Stack server, if different.
---------
Signed-off-by: Ben Browning <bbrownin@redhat.com>
# What does this PR do?
This PR adds unit tests for the NVIDIA Safety provider implementation.
[//]: # (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.*]
1. Ran `./scripts/unit-tests.sh
tests/unit/providers/nvidia/test_safety.py` from the root of the
project. Verified tests pass.
```
tests/unit/providers/nvidia/test_safety.py::TestNVIDIASafetyAdapter::test_init_nemo_guardrails Initializing NVIDIASafetyAdapter(http://nemo.test)...
PASSED
tests/unit/providers/nvidia/test_safety.py::TestNVIDIASafetyAdapter::test_init_nemo_guardrails_invalid_temperature Initializing NVIDIASafetyAdapter(http://nemo.test)...
PASSED
tests/unit/providers/nvidia/test_safety.py::TestNVIDIASafetyAdapter::test_register_shield_with_valid_id Initializing NVIDIASafetyAdapter(http://nemo.test)...
PASSED
tests/unit/providers/nvidia/test_safety.py::TestNVIDIASafetyAdapter::test_register_shield_without_id Initializing NVIDIASafetyAdapter(http://nemo.test)...
PASSED
tests/unit/providers/nvidia/test_safety.py::TestNVIDIASafetyAdapter::test_run_shield_allowed Initializing NVIDIASafetyAdapter(http://nemo.test)...
PASSED
tests/unit/providers/nvidia/test_safety.py::TestNVIDIASafetyAdapter::test_run_shield_blocked Initializing NVIDIASafetyAdapter(http://nemo.test)...
PASSED
tests/unit/providers/nvidia/test_safety.py::TestNVIDIASafetyAdapter::test_run_shield_http_error Initializing NVIDIASafetyAdapter(http://nemo.test)...
PASSED
tests/unit/providers/nvidia/test_safety.py::TestNVIDIASafetyAdapter::test_run_shield_not_found Initializing NVIDIASafetyAdapter(http://nemo.test)...
PASSED
```
[//]: # (## Documentation)
---------
Co-authored-by: Jash Gulabrai <jgulabrai@nvidia.com>
# What does this PR do?
## Test Plan
(myenv) ➜ llama-stack python tests/verifications/generate_report.py
--providers fireworks,together,openai --run-tests
# What does this PR do?
- provider and their models now live in config.yaml
- better distinguish different cases within a test
- add model key to surface provider's model_id
- include example command to rerun single test case
## Test Plan
<img width="1173" alt="image"
src="https://github.com/user-attachments/assets/b414baf0-c768-451f-8c3b-c2905cf36fac"
/>
# What does this PR do?
Fixes issue #1537 that causes "500 Internal Server Error" when
unregistering a toolgroup
# (Closes#1537 )
## Test Plan
```console
$ pytest -s -v tests/integration/tool_runtime/test_registration.py --stack-config=ollama --env INFERENCE_MODEL="meta-llama/Llama-3.2-3B-Instruct"
INFO 2025-03-14 21:15:03,999 tests.integration.conftest:41 tests: Setting DISABLE_CODE_SANDBOX=1 for macOS
/opt/homebrew/lib/python3.10/site-packages/pytest_asyncio/plugin.py:207: PytestDeprecationWarning: The configuration option "asyncio_default_fixture_loop_scope" is unset.
The event loop scope for asynchronous fixtures will default to the fixture caching scope. Future versions of pytest-asyncio will default the loop scope for asynchronous fixtures to function scope. Set the default fixture loop scope explicitly in order to avoid unexpected behavior in the future. Valid fixture loop scopes are: "function", "class", "module", "package", "session"
warnings.warn(PytestDeprecationWarning(_DEFAULT_FIXTURE_LOOP_SCOPE_UNSET))
===================================================== test session starts =====================================================
platform darwin -- Python 3.10.16, pytest-8.3.5, pluggy-1.5.0 -- /opt/homebrew/opt/python@3.10/bin/python3.10
cachedir: .pytest_cache
rootdir: /Users/paolo/Projects/aiplatform/llama-stack
configfile: pyproject.toml
plugins: asyncio-0.25.3, anyio-4.8.0
asyncio: mode=strict, asyncio_default_fixture_loop_scope=None
collected 1 item
tests/integration/tool_runtime/test_registration.py::test_register_and_unregister_toolgroup[None-None-None-None-None] INFO 2025-03-14 21:15:04,478 llama_stack.providers.remote.inference.ollama.ollama:75 inference: checking
connectivity to Ollama at `http://localhost:11434`...
INFO 2025-03-14 21:15:05,350 llama_stack.providers.remote.inference.ollama.ollama:294 inference: Pulling embedding
model `all-minilm:latest` if necessary...
INFO: Started server process [78391]
INFO: Waiting for application startup.
INFO: Application startup complete.
INFO: Uvicorn running on http://0.0.0.0:8000 (Press CTRL+C to quit)
INFO: 127.0.0.1:57424 - "GET /sse HTTP/1.1" 200 OK
INFO: 127.0.0.1:57434 - "GET /sse HTTP/1.1" 200 OK
INFO 2025-03-14 21:15:16,129 mcp.client.sse:51 uncategorized: Connecting to SSE endpoint: http://localhost:8000/sse
INFO: 127.0.0.1:57445 - "GET /sse HTTP/1.1" 200 OK
INFO 2025-03-14 21:15:16,146 mcp.client.sse:71 uncategorized: Received endpoint URL:
http://localhost:8000/messages/?session_id=c5b6fc01f8dc4b5e80e38eb1c1b22a9b
INFO 2025-03-14 21:15:16,147 mcp.client.sse:140 uncategorized: Starting post writer with endpoint URL:
http://localhost:8000/messages/?session_id=c5b6fc01f8dc4b5e80e38eb1c1b22a9b
INFO: 127.0.0.1:57447 - "POST /messages/?session_id=c5b6fc01f8dc4b5e80e38eb1c1b22a9b HTTP/1.1" 202 Accepted
INFO: 127.0.0.1:57447 - "POST /messages/?session_id=c5b6fc01f8dc4b5e80e38eb1c1b22a9b HTTP/1.1" 202 Accepted
INFO: 127.0.0.1:57447 - "POST /messages/?session_id=c5b6fc01f8dc4b5e80e38eb1c1b22a9b HTTP/1.1" 202 Accepted
INFO 2025-03-14 21:15:16,155 mcp.server.lowlevel.server:535 uncategorized: Processing request of type
ListToolsRequest
PASSED
=============================================== 1 passed, 4 warnings in 12.17s ================================================
```
---------
Signed-off-by: Paolo Dettori <dettori@us.ibm.com>
# What does this PR do?
Providers that live outside of the llama-stack codebase are now
supported.
A new property `external_providers_dir` has been added to the main
config and can be configured as follow:
```
external_providers_dir: /etc/llama-stack/providers.d/
```
Where the expected structure is:
```
providers.d/
inference/
custom_ollama.yaml
vllm.yaml
vector_io/
qdrant.yaml
```
Where `custom_ollama.yaml` is:
```
adapter:
adapter_type: custom_ollama
pip_packages: ["ollama", "aiohttp"]
config_class: llama_stack_ollama_provider.config.OllamaImplConfig
module: llama_stack_ollama_provider
api_dependencies: []
optional_api_dependencies: []
```
Obviously the package must be installed on the system, here is the
`llama_stack_ollama_provider` example:
```
$ uv pip show llama-stack-ollama-provider
Using Python 3.10.16 environment at: /Users/leseb/Documents/AI/llama-stack/.venv
Name: llama-stack-ollama-provider
Version: 0.1.0
Location: /Users/leseb/Documents/AI/llama-stack/.venv/lib/python3.10/site-packages
Editable project location: /private/var/folders/mq/rnm5w_7s2d3fxmtkx02knvhm0000gn/T/tmp.ZBHU5Ezxg4/ollama/llama-stack-ollama-provider
Requires:
Required-by:
```
Closes: https://github.com/meta-llama/llama-stack/issues/658
Signed-off-by: Sébastien Han <seb@redhat.com>
Move the test_context.py under the main tests directory, and fix the
code.
The problem was that the function captures the initial values of the
context variables and then restores those same initial values before
each iteration. This means that any modifications made to the context
variables during iteration are lost when the next iteration starts.
Error was:
```
====================================================== FAILURES =======================================================
______________________________________ test_preserve_contexts_across_event_loops ______________________________________
@pytest.mark.asyncio
async def test_preserve_contexts_across_event_loops():
"""
Test that context variables are preserved across event loop boundaries with nested generators.
This simulates the real-world scenario where:
1. A new event loop is created for each streaming request
2. The async generator runs inside that loop
3. There are multiple levels of nested generators
4. Context needs to be preserved across these boundaries
"""
# Create context variables
request_id = ContextVar("request_id", default=None)
user_id = ContextVar("user_id", default=None)
# Set initial values
# Results container to verify values across thread boundaries
results = []
# Inner-most generator (level 2)
async def inner_generator():
# Should have the context from the outer scope
yield (1, request_id.get(), user_id.get())
# Modify one context variable
user_id.set("user-modified")
# Should reflect the modification
yield (2, request_id.get(), user_id.get())
# Middle generator (level 1)
async def middle_generator():
inner_gen = inner_generator()
# Forward the first yield from inner
item = await inner_gen.__anext__()
yield item
# Forward the second yield from inner
item = await inner_gen.__anext__()
yield item
request_id.set("req-modified")
# Add our own yield with both modified variables
yield (3, request_id.get(), user_id.get())
# Function to run in a separate thread with a new event loop
def run_in_new_loop():
# Create a new event loop for this thread
loop = asyncio.new_event_loop()
asyncio.set_event_loop(loop)
try:
# Outer generator (runs in the new loop)
async def outer_generator():
request_id.set("req-12345")
user_id.set("user-6789")
# Wrap the middle generator
wrapped_gen = preserve_contexts_async_generator(middle_generator(), [request_id, user_id])
# Process all items from the middle generator
async for item in wrapped_gen:
# Store results for verification
results.append(item)
# Run the outer generator in the new loop
loop.run_until_complete(outer_generator())
finally:
loop.close()
# Run the generator chain in a separate thread with a new event loop
with ThreadPoolExecutor(max_workers=1) as executor:
future = executor.submit(run_in_new_loop)
future.result() # Wait for completion
# Verify the results
assert len(results) == 3
# First yield should have original values
assert results[0] == (1, "req-12345", "user-6789")
# Second yield should have modified user_id
assert results[1] == (2, "req-12345", "user-modified")
# Third yield should have both modified values
> assert results[2] == (3, "req-modified", "user-modified")
E AssertionError: assert (3, 'req-modified', 'user-6789') == (3, 'req-modified', 'user-modified')
E
E At index 2 diff: 'user-6789' != 'user-modified'
E
E Full diff:
E (
E 3,
E 'req-modified',
E - 'user-modified',
E + 'user-6789',
E )
tests/unit/distribution/test_context.py:155: AssertionError
-------------------------------------------------- Captured log call --------------------------------------------------
ERROR asyncio:base_events.py:1758 Task was destroyed but it is pending!
task: <Task pending name='Task-7' coro=<<async_generator_athrow without __name__>()>>
================================================== warnings summary ===================================================
.venv/lib/python3.10/site-packages/pydantic/fields.py:1042
/Users/leseb/Documents/AI/llama-stack/.venv/lib/python3.10/site-packages/pydantic/fields.py:1042: PydanticDeprecatedSince20: Using extra keyword arguments on `Field` is deprecated and will be removed. Use `json_schema_extra` instead. (Extra keys: 'contentEncoding'). Deprecated in Pydantic V2.0 to be removed in V3.0. See Pydantic V2 Migration Guide at https://errors.pydantic.dev/2.10/migration/
warn(
-- Docs: https://docs.pytest.org/en/stable/how-to/capture-warnings.html
=============================================== short test summary info ===============================================
FAILED tests/unit/distribution/test_context.py::test_preserve_contexts_across_event_loops - AssertionError: assert (3, 'req-modified', 'user-6789') == (3, 'req-modified', 'user-modified')
At index 2 diff: 'user-6789' != 'user-modified'
Full diff:
(
3,
'req-modified',
- 'user-modified',
+ 'user-6789',
)
```
[//]: # (## Documentation)
Signed-off-by: Sébastien Han <seb@redhat.com>
# What does this PR do?
## Test Plan
export MODEL=accounts/fireworks/models/llama4-scout-instruct-basic;
LLAMA_STACK_CONFIG=verification pytest -s -v tests/integration/inference
--vision-model $MODEL --text-model $MODEL
# What does this PR do?
Move around bits. This makes the copies from llama-models _much_ easier
to maintain and ensures we don't entangle meta-reference specific
tidbits into llama-models code even by accident.
Also, kills the meta-reference-quantized-gpu distro and rolls
quantization deps into meta-reference-gpu.
## Test Plan
```
LLAMA_MODELS_DEBUG=1 \
with-proxy llama stack run meta-reference-gpu \
--env INFERENCE_MODEL=meta-llama/Llama-4-Scout-17B-16E-Instruct \
--env INFERENCE_CHECKPOINT_DIR=<DIR> \
--env MODEL_PARALLEL_SIZE=4 \
--env QUANTIZATION_TYPE=fp8_mixed
```
Start a server with and without quantization. Point integration tests to
it using:
```
pytest -s -v tests/integration/inference/test_text_inference.py \
--stack-config http://localhost:8321 --text-model meta-llama/Llama-4-Scout-17B-16E-Instruct
```
Running full Tool Calling required some updates to work e2e.
- Remove `python_start` and `python_end` tags
- Tool Call messages and Tool Resposne messages should end with
`<|eom|>`
- System prompt needed updates
```
You are a helpful assisant who can can answer general questions or invoke tools when necessary.
In addition to tool calls, you should also augment your responses by using the tool outputs.
```
### Test Plan
- Start server with meta-reference
```
LLAMA_STACK_DISABLE_VERSION_CHECK=1 LLAMA_MODELS_DEBUG=1 INFERENCE_MODEL=meta-llama/$MODEL llama stack run meta-reference-gpu
```
- Added **NEW** tests with 5 test cases for multi-turn tool calls
```
pytest -s -v --stack-config http://localhost:8321 tests/integration/inference/test_text_inference.py --text-model meta-llama/Llama-4-Scout-17B-16E-Instruct
```
- Also verified all vision and agent tests pass
Previously, the integration tests started the server, but never really
used it because `--stack-config=ollama` uses the ollama template and the
inline "llama stack as library" client, not the HTTP client.
This PR makes sure we test it both ways.
We also add agents tests to the mix.
## Test Plan
GitHub
---------
Signed-off-by: Sébastien Han <seb@redhat.com>
Co-authored-by: Sébastien Han <seb@redhat.com>
# What does this PR do?
* Added `--text-model` in example command.
* Added link to integration tests instruction and a note on specifying
models.
This is to avoid confusion when all tests are skipped because no model
is provided.
Signed-off-by: Yuan Tang <terrytangyuan@gmail.com>
# What does this PR do?
https://github.com/meta-llama/llama-stack/pull/1828 removed
__root_span__ attribute which is still needed
## Test Plan
added telemetry integration test
LLAMA_STACK_CONFIG=http://localhost:5001 pytest -s -v
tests/integration/telemetry --safety-shield meta-llama/Llama-Guard-3-8B
--text-model accounts/fireworks/models/llama-v3p3-70b-instruct
