# What does this PR do?
remove unused chat_completion implementations
vllm features ported -
- requires max_tokens be set, use config value
- set tool_choice to none if no tools provided
## Test Plan
ci
This is a sweeping change to clean up some gunk around our "Tool"
definitions.
First, we had two types `Tool` and `ToolDef`. The first of these was a
"Resource" type for the registry but we had stopped registering tools
inside the Registry long back (and only registered ToolGroups.) The
latter was for specifying tools for the Agents API. This PR removes the
former and adds an optional `toolgroup_id` field to the latter.
Secondly, as pointed out by @bbrowning in
https://github.com/llamastack/llama-stack/pull/3003#issuecomment-3245270132,
we were doing a lossy conversion from a full JSON schema from the MCP
tool specification into our ToolDefinition to send it to the model.
There is no necessity to do this -- we ourselves aren't doing any
execution at all but merely passing it to the chat completions API which
supports this. By doing this (and by doing it poorly), we encountered
limitations like not supporting array items, or not resolving $refs,
etc.
To fix this, we replaced the `parameters` field by `{ input_schema,
output_schema }` which can be full blown JSON schemas.
Finally, there were some types in our llama-related chat format
conversion which needed some cleanup. We are taking this opportunity to
clean those up.
This PR is a substantial breaking change to the API. However, given our
window for introducing breaking changes, this suits us just fine. I will
be landing a concurrent `llama-stack-client` change as well since API
shapes are changing.
# 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?
- 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?
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?
*Add dynamic authentication token forwarding support for vLLM provider*
This enables per-request authentication tokens for vLLM providers,
supporting use cases like RAG operations where different requests may
need different authentication tokens. The implementation follows the
same pattern as other providers like Together AI, Fireworks, and
Passthrough.
- Add LiteLLMOpenAIMixin that manages the vllm_api_token properly
Usage:
- Static: VLLM_API_TOKEN env var or config.api_token
- Dynamic: X-LlamaStack-Provider-Data header with vllm_api_token
All existing functionality is preserved while adding new dynamic
capabilities.
<!-- Provide a short summary of what this PR does and why. Link to
relevant issues if applicable. -->
<!-- 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.* -->
```
curl -X POST "http://localhost:8000/v1/chat/completions" -H "Authorization: Bearer my-dynamic-token" \
-H "X-LlamaStack-Provider-Data: {\"vllm_api_token\": \"Bearer my-dynamic-token\", \"vllm_url\": \"http://dynamic-server:8000\"}" \
-H "Content-Type: application/json" \
-d '{"model": "llama-3.1-8b", "messages": [{"role": "user", "content": "Hello!"}]}'
```
---------
Signed-off-by: Akram Ben Aissi <akram.benaissi@gmail.com>
# What does this PR do?
update the async detection test for vllm
- remove a network access from unit tests
- remove direct logging use
the idea behind the test is to mock inference w/ a sleep, initiate
concurrent inference calls, verify the total execution time is close to
the sleep time. in a non-async env the total time would be closer to
sleep * num concurrent calls.
## Test Plan
ci
# 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?
update Groq inference provider to use OpenAIMixin for openai-compat
endpoints
changes on api.groq.com -
- json_schema is now supported for specific models, see
https://console.groq.com/docs/structured-outputs#supported-models
- response_format with streaming is now supported for models that
support response_format
- groq no longer returns a 400 error if tools are provided and
tool_choice is not "required"
## Test Plan
```
$ GROQ_API_KEY=... uv run llama stack build --image-type venv --providers inference=remote::groq --run
...
$ LLAMA_STACK_CONFIG=http://localhost:8321 uv run --group test pytest -v -ra --text-model groq/llama-3.3-70b-versatile tests/integration/inference/test_openai_completion.py -k 'not store'
...
SKIPPED [3] tests/integration/inference/test_openai_completion.py:44: Model groq/llama-3.3-70b-versatile hosted by remote::groq doesn't support OpenAI completions.
SKIPPED [3] tests/integration/inference/test_openai_completion.py:94: Model groq/llama-3.3-70b-versatile hosted by remote::groq doesn't support vllm extra_body parameters.
SKIPPED [4] tests/integration/inference/test_openai_completion.py:73: Model groq/llama-3.3-70b-versatile hosted by remote::groq doesn't support n param.
SKIPPED [1] tests/integration/inference/test_openai_completion.py💯 Model groq/llama-3.3-70b-versatile hosted by remote::groq doesn't support chat completion calls with base64 encoded files.
======================= 8 passed, 11 skipped, 8 deselected, 2 warnings in 5.13s ========================
```
---------
Co-authored-by: raghotham <rsm@meta.com>
# What does this PR do?
Improved bedrock provider config to read from environment variables like
AWS_ACCESS_KEY_ID. Updated all
fields to use default_factory with lambda patterns like the nvidia
provider does.
Now the environment variables work as documented.
Closes#3305
## Test Plan
Ran the new bedrock config tests:
```bash
python -m pytest tests/unit/providers/inference/bedrock/test_config.py
-v
Verified existing provider tests still work:
python -m pytest tests/unit/providers/test_configs.py -v
# 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>
# 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?
- Add base_url field to OpenAIConfig with default
"https://api.openai.com/v1"
- Update sample_run_config to support OPENAI_BASE_URL environment
variable
- Modify get_base_url() to return configured base_url instead of
hardcoded value
- Add comprehensive test suite covering:
- Default base URL behavior
- Custom base URL from config
- Environment variable override
- Config precedence over environment variables
- Client initialization with configured URL
- Model availability checks using configured URL
This enables users to configure custom OpenAI-compatible API endpoints
via environment variables or configuration files.
Closes#2910
## Test Plan
run unit tests
# 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
# What does this PR do?
this blocks network access for all `tests/unit/` tests.
`tests/integration/` are untouched.
it also introduces an `allow_network` marker to explicitly allow network
access.
## Test Plan
`./scripts/unit-tests.sh`
# What does this PR do?
Some of our inference providers support passthrough authentication via
`x-llamastack-provider-data` header values. This fixes the providers
that support passthrough auth to not cache their clients to the backend
providers (mostly OpenAI client instances) so that the client connecting
to Llama Stack has to provide those auth values on each and every
request.
## Test Plan
I added some unit tests to ensure we're not caching clients across
requests for all the fixed providers in this PR.
```
uv run pytest -sv tests/unit/providers/inference/test_inference_client_caching.py
```
I also ran some of our OpenAI compatible API integration tests for each
of the changed providers, just to ensure they still work. Note that
these providers don't actually pass all these tests (for unrelated
reasons due to quirks of the Groq and Together SaaS services), but
enough of the tests passed to confirm the clients are still working as
intended.
### Together
```
ENABLE_TOGETHER="together" \
uv run llama stack run llama_stack/templates/starter/run.yaml
LLAMA_STACK_CONFIG=http://localhost:8321 \
uv run pytest -sv \
tests/integration/inference/test_openai_completion.py \
--text-model "together/meta-llama/Llama-3.1-8B-Instruct"
```
### OpenAI
```
ENABLE_OPENAI="openai" \
uv run llama stack run llama_stack/templates/starter/run.yaml
LLAMA_STACK_CONFIG=http://localhost:8321 \
uv run pytest -sv \
tests/integration/inference/test_openai_completion.py \
--text-model "openai/gpt-4o-mini"
```
### Groq
```
ENABLE_GROQ="groq" \
uv run llama stack run llama_stack/templates/starter/run.yaml
LLAMA_STACK_CONFIG=http://localhost:8321 \
uv run pytest -sv \
tests/integration/inference/test_openai_completion.py \
--text-model "groq/meta-llama/Llama-3.1-8B-Instruct"
```
---------
Signed-off-by: Ben Browning <bbrownin@redhat.com>
# What does this PR do?
previously, developers who ran `./scripts/unit-tests.sh` would get
`asyncio-mode=auto`, which meant `@pytest.mark.asyncio` and
`@pytest_asyncio.fixture` were redundent. developers who ran `pytest`
directly would get pytest's default (strict mode), would run into errors
leading them to add `@pytest.mark.asyncio` / `@pytest_asyncio.fixture`
to their code.
with this change -
- `asyncio_mode=auto` is included in `pyproject.toml` making behavior
consistent for all invocations of pytest
- removes all redundant `@pytest_asyncio.fixture` and
`@pytest.mark.asyncio`
- for good measure, requires `pytest>=8.4` and `pytest-asyncio>=1.0`
## Test Plan
- `./scripts/unit-tests.sh`
- `uv run pytest tests/unit`
# What does this PR do?
<!-- Provide a short summary of what this PR does and why. Link to
relevant issues if applicable. -->
To add health status check for remote VLLM
<!-- 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.* -->
PR includes the unit test to test the added health check implementation
feature.
# What does this PR do?
The remote-vllm `test_chat_completion_doesnt_block_event_loop` unit test
was often failing for me on a Mac with a `httpx.ReadError`. I traced
this back to the swap to the `AsyncOpenAI` client in the remote-vllm
provider as where this started, and it looks like the async client needs
a bit more accurate HTTP request handling from our mock server.
So, this fixes that unit test to send proper Content-Type and
Content-Length headers which makes the `AsyncOpenAI` client happier on
Macs.
## Test Plan
All the test_remote_vllm.py unit tests consistently pass for me on a Mac
now, without any flaking in the event loop one.
`pytest -s -v tests/unit/providers/inference/test_remote_vllm.py`
Signed-off-by: Ben Browning <bbrownin@redhat.com>
# What does this PR do?
This fixes an issue in how we used the tool_call_buf from streaming tool
calls in the remote-vllm provider where it would end up concatenating
parameters from multiple different tool call results instead of
aggregating the results from each tool call separately.
It also fixes an issue found while digging into that where we were
accidentally mixing the json string form of tool call parameters with
the string representation of the python form, which mean we'd end up
with single quotes in what should be double-quoted json strings.
Closes#1120
## Test Plan
The following tests are now passing 100% for the remote-vllm provider,
where some of the test_text_inference were failing before this change:
```
VLLM_URL="http://localhost:8000/v1" INFERENCE_MODEL="RedHatAI/Llama-4-Scout-17B-16E-Instruct-FP8-dynamic" LLAMA_STACK_CONFIG=remote-vllm python -m pytest -v tests/integration/inference/test_text_inference.py --text-model "RedHatAI/Llama-4-Scout-17B-16E-Instruct-FP8-dynamic"
VLLM_URL="http://localhost:8000/v1" INFERENCE_MODEL="RedHatAI/Llama-4-Scout-17B-16E-Instruct-FP8-dynamic" LLAMA_STACK_CONFIG=remote-vllm python -m pytest -v tests/integration/inference/test_vision_inference.py --vision-model "RedHatAI/Llama-4-Scout-17B-16E-Instruct-FP8-dynamic"
```
All but one of the agent tests are passing (including the multi-tool
one). See the PR at https://github.com/vllm-project/vllm/pull/17917 and
a gist at
https://gist.github.com/bbrowning/4734240ce96b4264340caa9584e47c9e for
changes needed there, which will have to get made upstream in vLLM.
Agent tests:
```
VLLM_URL="http://localhost:8000/v1" INFERENCE_MODEL="RedHatAI/Llama-4-Scout-17B-16E-Instruct-FP8-dynamic" LLAMA_STACK_CONFIG=remote-vllm python -m pytest -v tests/integration/agents/test_agents.py --text-model "RedHatAI/Llama-4-Scout-17B-16E-Instruct-FP8-dynamic"
````
---------
Signed-off-by: Ben Browning <bbrownin@redhat.com>
# What does this PR do?
Closes#2113.
Closes#1783.
Fixes a bug in handling the end of tool execution request stream where
no `finish_reason` is provided by the model.
## Test Plan
1. Ran existing unit tests
2. Added a dedicated test verifying correct behavior in this edge case
3. Ran the code snapshot from #2113
[//]: # (## Documentation)
# What does this PR do?
Closes#2111.
Fixes an error causing Llama Stack to just return `<tool_call>` and
complete the turn without actually executing the tool. See the issue
description for more detail.
## Test Plan
1) Ran existing unit tests
2) Added a dedicated test verifying correct behavior in this edge case
3) Ran the code snapshot from #2111
# What does this PR do?
The goal of this PR is code base modernization.
Schema reflection code needed a minor adjustment to handle UnionTypes
and collections.abc.AsyncIterator. (Both are preferred for latest Python
releases.)
Note to reviewers: almost all changes here are automatically generated
by pyupgrade. Some additional unused imports were cleaned up. The only
change worth of note can be found under `docs/openapi_generator` and
`llama_stack/strong_typing/schema.py` where reflection code was updated
to deal with "newer" types.
Signed-off-by: Ihar Hrachyshka <ihar.hrachyshka@gmail.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>
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?
DocVQA asks model to look a a picture, then answer a question given in
text, with a text answer by text information in the picture. these
questions often require understanding of relative positions of texts
within the picture.
original dataset is defined in the "Task1" of
https://www.docvqa.org/datasets
## Test Plan
setup llama server with
```
llama stack run ./llama_stack/templates/open-benchmark/run.yaml
```
then send traffic:
```
llama-stack-client eval run-benchmark "meta-reference-docvqa" --model-id meta-llama/Llama-3.3-70B-Instruct --output-dir /tmp/gpqa --num-examples 200
```
# What does this PR do?
This avoids flaky timeout issue observed in CI builds, e.g.
3891286596
## Test Plan
Ran multiple times and pass consistently.
Signed-off-by: Yuan Tang <terrytangyuan@gmail.com>
# What does this PR do?
This switches from an OpenAI client to the AsyncOpenAI client in the
remote vllm provider. The main benefit of this is that instead of each
client call being a blocking operation that was blocking our server
event loop, the client calls are now async operations that do not block
the event loop.
The actual fix is quite simple and straightforward. Creating a reliable
reproducer of this with a unit test that verifies we were blocking the
event loop before and are not blocking it any longer was a bit harder.
Some other inference providers have this same issue, so we may want to
make that simple delayed http server a bit more generic and pull it into
a common place as other inference providers get fixed.
(Closes#1457)
## Test Plan
I verified the unit tests and test_text_inference tests pass with this
change like below:
```
python -m pytest -v tests/unit
```
```
VLLM_URL="http://localhost:8000/v1" \
INFERENCE_MODEL="meta-llama/Llama-3.2-3B-Instruct" \
LLAMA_STACK_CONFIG=remote-vllm \
python -m pytest -v -s \
tests/integration/inference/test_text_inference.py \
--text-model "meta-llama/Llama-3.2-3B-Instruct"
```
Signed-off-by: Ben Browning <bbrownin@redhat.com>
# What does this PR do?
This gracefully handles the case where the vLLM server responded to a
completion request with no choices, which can happen in certain vLLM
error situations. Previously, we'd error out with a stack trace about a
list index out of range. Now, we just log a warning to the user and move
past any chunks with an empty choices list.
A specific example of the type of stack trace this fixes:
```
File "/app/llama-stack-source/llama_stack/providers/remote/inference/vllm/vllm.py", line 170, in _process_vllm_chat_completion_stream_response
choice = chunk.choices[0]
~~~~~~~~~~~~~^^^
IndexError: list index out of range
```
Now, instead of erroring out with that stack trace, we log a warning
that vLLM failed to generate any completions and alert the user to check
the vLLM server logs for details.
This is related to #1277 and addresses the stack trace shown in that
issue, although does not in and of itself change the functional behavior
of vLLM tool calling.
## Test Plan
As part of this fix, I added new unit tests to trigger this same error
and verify it no longer happens. That is
`test_process_vllm_chat_completion_stream_response_no_choices` in the
new `tests/unit/providers/inference/test_remote_vllm.py`. I also added a
couple of more tests to trigger and verify the last couple of remote
vllm provider bug fixes - specifically a test for #1236 (builtin tool
calling) and #1325 (vLLM <= v0.6.3).
This required fixing the signature of
`_process_vllm_chat_completion_stream_response` to accept the actual
type of chunks it was getting passed - specifically changing from our
openai_compat `OpenAICompatCompletionResponse` to
`openai.types.chat.chat_completion_chunk.ChatCompletionChunk`. It was
not actually getting passed `OpenAICompatCompletionResponse` objects
before, and was using attributes that didn't exist on those objects. So,
the signature now matches the type of object it's actually passed.
Run these new unit tests like this:
```
pytest tests/unit/providers/inference/test_remote_vllm.py
```
Additionally, I ensured the existing `test_text_inference.py` tests
passed via:
```
VLLM_URL="http://localhost:8000/v1" \
INFERENCE_MODEL="meta-llama/Llama-3.2-3B-Instruct" \
LLAMA_STACK_CONFIG=remote-vllm \
python -m pytest -v tests/integration/inference/test_text_inference.py \
--inference-model "meta-llama/Llama-3.2-3B-Instruct" \
--vision-inference-model ""
```
Signed-off-by: Ben Browning <bbrownin@redhat.com>
