jeff3071 opened a new pull request, #71437:
URL: https://github.com/apache/airflow/pull/71437
<!-- SPDX-License-Identifier: Apache-2.0
https://www.apache.org/licenses/LICENSE-2.0 -->
<!--
Thank you for contributing!
Please provide above a brief description of the changes made in this pull
request.
Write a good git commit message following this guide:
https://chris.beams.io/posts/git-commit/
Please make sure that your code changes are covered with tests.
And in case of new features or big changes remember to adjust the
documentation.
For user-facing UI changes, please attach before/after screenshots (or a
short
screen recording) so reviewers can assess the visual impact.
Feel free to ping (in general) for the review if you do not see reaction for
a few days
(72 Hours is the minimum reaction time you can expect from volunteers) - we
sometimes miss notifications.
In case of an existing issue, reference it using one of the following:
* closes: #ISSUE
* related: #ISSUE
-->
## Why
`PydanticAIHook` supports Airflow-managed credentials and custom base URLs
for LLMs, including OpenAI-compatible self-hosted endpoints, but it does not
expose pydantic-ai's embedding API. Users cannot create a pydantic-ai
`Embedder`.
## What
- Add `embed_model_id` and `extra["embed_model"]` resolution to
`PydanticAIHook`.
- Add a cached `get_embedder()` that returns a pydantic-ai `Embedder`
configured with the connection's credentials and base URL.
- Share provider-factory creation between LLM and embedding model resolution
so provider-specific hooks keep their existing authentication mapping.
## Testing
- Manual Airflow UI verification against a vLLM-hosted
`Qwen/Qwen3-Embedding-0.6B` OpenAI-compatible endpoint:
- one query vector and two document vectors returned;
- all vectors dimension is 1024;
`serve command`
```bash
vllm serve Qwen/Qwen3-Embedding-0.6B --runner pooling --host 0.0.0.0
--port 8000 --api-key EMPTY --max-model-len 1024
--gpu-memory-utilization 0.8
```
<!-- Attach the Airflow task success/XCom screenshot here. -->
<img width="905" height="313" alt="test"
src="https://github.com/user-attachments/assets/9c989f8c-1496-4560-9a8b-78e3e92cd4ee";
/>
test dag
```python
from __future__ import annotations
import math
from airflow.providers.common.ai.hooks.pydantic_ai import PydanticAIHook
from airflow.sdk import dag, task
CONN_ID = "pydanticai_qwen_embedding"
EXPECTED_DIMENSION = 1024
@dag(
dag_id="qwen3_self_hosted_embedding",
schedule=None,
catchup=False,
tags=["manual", "pydantic-ai", "embedding", "self-hosted"],
)
def qwen3_self_hosted_embedding():
@task
def verify_embedding() -> dict[str, int | str | bool]:
hook = PydanticAIHook(llm_conn_id=CONN_ID)
embedder = hook.get_embedder()
query_result = embedder.embed_query_sync("How does Apache Airflow
orchestrate workflows?")
document_result = embedder.embed_documents_sync(
[
"Apache Airflow is a platform for developing and monitoring
batch workflows.",
"Qwen3 Embedding converts text into dense vector
representations.",
]
)
if len(query_result.embeddings) != 1 or
len(document_result.embeddings) != 2:
raise ValueError(
"Unexpected vector count: "
f"query={len(query_result.embeddings)},
documents={len(document_result.embeddings)}"
)
vectors = [*query_result.embeddings, *document_result.embeddings]
dimensions = {len(vector) for vector in vectors}
if dimensions != {EXPECTED_DIMENSION}:
raise ValueError(f"Expected {EXPECTED_DIMENSION}-dimension
vectors, got {sorted(dimensions)}")
if not all(math.isfinite(value) for vector in vectors for value in
vector):
raise ValueError("Embedding response contains non-finite
values.")
return {
"model": query_result.model_name,
"provider": query_result.provider_name,
"dimension": next(iter(dimensions)),
"query_vectors": len(query_result.embeddings),
"document_vectors": len(document_result.embeddings),
"embedder_cached": embedder is hook.get_embedder(),
}
verify_embedding()
qwen3_self_hosted_embedding()
```
`log`
```
/root/airflow/logs/dag_id=qwen3_self_hosted_embedding/run_id=manual__2026-08-11T12:49:12.574
572+00:00/task_id=verify_embedding/attempt=1.log
::endgroup::
[2026-08-11T12:49:13.311809Z] INFO - ::group::Pre Execute
Task Identity ti_id=019ff0de-ab8f-7043-84dc-098e5407c3c5
dag_id=qwen3_self_hosted_embedding task_id=verify_embedding
run_id=manual__2026-08-11T12:49:12.574572+00:00 try_number=1 map_index=-1
[2026-08-11T12:49:13.363284Z] INFO - DAG bundles loaded: dags-folder
[2026-08-11T12:49:13.365408Z] INFO - Filling up the DagBag from
/files/dags/manual_qwen3_self_hosted_embedding.py
[2026-08-11T12:49:15.292911Z] INFO - Worker startup parse complete
bundle_name=dags-folder bundle_version=null
dag_file=manual_qwen3_self_hosted_embedding.py bundle_prepare_ms=3
dag_file_parse_ms=1928
[2026-08-11T12:49:15.326585Z] INFO - ::endgroup::
[2026-08-11T12:49:15.352060Z] INFO - Using explicit credentials for provider
with embedding model 'openai:Qwen/Qwen3-Embedding-0.6B': ['api_key', 'base_url']
[2026-08-11T12:49:17.976642Z] INFO - Done. Returned value was: {'model':
'Qwen/Qwen3-Embedding-0.6B', 'provider': 'openai', 'dimension': 1024,
'query_vectors': 1, 'document_vectors': 2, 'embedder_cached': True}
[2026-08-11T12:49:17.976832Z] INFO - ::group::Post Execute
[2026-08-11T12:49:17.976973Z] INFO - Pushing xcom
ti=RuntimeTaskInstance(id=UUID('019ff0de-ab8f-7043-84dc-098e5407c3c5'),
task_id='verify_embedding', dag_id='qwen3_self_hosted_embedding',
run_id='manual__2026-08-11T12:49:12.574572+00:00', try_number=1,
dag_version_id=UUID('019ff0dd-ef97-7b8e-9edb-36664c8c2b3a'), map_index=-1,
hostname='86afabd35b10', context_carrier={'traceparent':
'00-a5a16744aec0bf462e7b31c73faa796d-b98449d8ef65dcfd-00'}, queue='default',
task=<Task(_PythonDecoratedOperator): verify_embedding>,
bundle_instance=LocalDagBundle(name=dags-folder), max_tries=0,
start_date=datetime.datetime(2026, 8, 11, 12, 49, 13, 344044,
tzinfo=datetime.timezone.utc), end_date=None, state=<TaskInstanceState.RUNNING:
'running'>, is_mapped=False, rendered_map_index=None, sentry_integration='')
[2026-08-11T12:49:18.094619Z] INFO - ::endgroup::
```
---
---
##### Was generative AI tooling used to co-author this PR?
<!--
If generative AI tooling has been used in the process of authoring this PR,
please
change below checkbox to `[X]` followed by the name of the tool, uncomment
the "Generated-by".
-->
- [X] Yes — Codex (GPT-5.6-sol)
Generated-by: Codex (GPT-5.6-sol) following [the
guidelines](https://github.com/apache/airflow/blob/main/contributing-docs/05_pull_requests.rst#gen-ai-assisted-contributions)
<!--
Generated-by: [Tool Name] following [the
guidelines](https://github.com/apache/airflow/blob/main/contributing-docs/05_pull_requests.rst#gen-ai-assisted-contributions)
-->
---
* Read the **[Pull Request
Guidelines](https://github.com/apache/airflow/blob/main/contributing-docs/05_pull_requests.rst#pull-request-guidelines)**
for more information. Note: commit author/co-author name and email in commits
become permanently public when merged.
* For fundamental code changes, an Airflow Improvement Proposal
([AIP](https://cwiki.apache.org/confluence/display/AIRFLOW/Airflow+Improvement+Proposals))
is needed.
* When adding dependency, check compliance with the [ASF 3rd Party License
Policy](https://www.apache.org/legal/resolved.html#category-x).
* For significant user-facing changes create newsfragment:
`{pr_number}.significant.rst`, in
[airflow-core/newsfragments](https://github.com/apache/airflow/tree/main/airflow-core/newsfragments).
You can add this file in a follow-up commit after the PR is created so you
know the PR number.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
