☁️ Build multimodal AI applications with cloud-native stack
3.25.1)Release time: 2024-04-10 14:39:02
This release contains 1 bug fix.
--provider Azure (#6163)When using --provider Azure, there was an error when using the input schema of an incorrect endpoint.
We would like to thank all contributors to this release:
3.25.0)Release time: 2024-04-02 13:02:31
This release contains 1 new feature.
Executors can now run as a FastAPI app in a Microsoft Azure container when being served with AZURE as the provider argument.
We would like to thank all contributors to this release:
3.24.1)Release time: 2024-03-20 09:35:52
This release contains 3 bug fixes.
float field (#6155)When serving a Flow with HTTP, serving failed because the model reconstructed in the Gateway interpreted the float field as an array. Now the model served in the Gateway uses the correct schema.
/invocations endpoint when serving with --provider SAGEMAKER (#6152)When exposing an Executor with SAGEMAKER as the provider argument, only the /invocations endpoint should be used. However, previously all other endpoints were also exposed. This is now fixed.
When using a strongly typed parameters argument with the Executor endpoint, even if the parameters model type had all fields with default values and not required, you still needed to pass an empty parameters={} field to the call when serving via HTTP. With this change, you no longer need to pass this empty dictionary.
We would like to thank all contributors to this release:
3.24.0)Release time: 2024-03-11 09:48:35
This release contains 1 new feature.
This release adds the --provider-endpoint flag, which can only be used in conjunction with the --provider flag. The --provider-endpoint flag takes as a parameter a named endpoint supported in the Executor and sets the endpoint of the provider to map to that Executor endpoint.
At present, SageMaker is the only available provider option, so using the parameters --provider SAGEMAKER --provider-endpoint <named endpoint> will make the SageMaker invocations endpoint map to the named endpoint of the Executor.
We would like to thank all contributors to this release:
3.23.5)Release time: 2024-02-29 14:36:02
This release contains 1 bug fix.
This release improves the SageMaker Batch Transform input data parsing logic so that data with more complex schema can be interpreted as expected.
We would like to thank all contributors to this release:
3.23.4)Release time: 2024-02-27 09:36:35
This release contains 1 bug fix.
dry_run endpoint (#6143)In prior releases, to assess Flow connectivity Jina automatically bound an Executor to a _dry_run_ endpoint via an object method. This release changes that to a class-unbound method, thus avoiding a cyclic reference in the inner requests object of the Executor.
We would like to thank all contributors to this release:
3.23.3)Release time: 2024-02-16 10:56:19
This release contains 1 bug fix and 1 documentation improvement.
Previously, nested document types with multiple DocList levels of nesting sometimes led to the schema exposed by the Gateway being incomplete because some references were lost.
Now schemas like this exposed through Flow work as expected:
class QuoteFile(BaseDoc):
quote_file_id: int
texts: DocList[TextDoc]
images: DocList[ImageDoc]
class SearchResult(BaseDoc):
results: DocList[QuoteFile]
3.23.2)Release time: 2023-12-14 15:28:24
This release contains 1 dependency update and 2 bug fixes.
The grpc version requirements have been updated to allow grpcio<=1.57.0.
An issue was identified when a dynamic batch raises an Exception that could affect some unrelated requests.
When using Deployment locally with multiple replicas, a load-balancing process is added in front of the replicas.
The load balancer assumed all GET requests to be streaming, but this may not be true for user-added FastAPI endpoints. We have fixed this assumption and now use the response type to determine if a request is streaming.
We would like to thank all contributors to this release:
3.23.1)Release time: 2023-12-01 09:24:22
This release contains 1 bug fix.
We fixed the dependency version with opentelemetry-exporter-prometheus to avoid using deprecated versions.
We would like to thank all contributors to this release:
3.23.0)Release time: 2023-11-20 10:02:37
This release contains 1 new feature and 1 bug fix.
When running a Deployment with HTTP locally without docker you can now extend the API by using extend_rest_interface, the same as with Flow.
import jina
from jina import Deployment
def extend_rest_function(app):
@app.get('/hello', tags=['My Extended APIs'])
async def foo():
return {'msg': 'hello world'}
return app
jina.helper.extend_rest_interface = extend_rest_function
d = Deployment(protocol='http', port=8080)
with d:
response = requests.get(f'http://localhost:8080/hello')
assert response.status_code == 200
assert response.json() == {'msg': 'hello world'}
There was an issue when using Jina to do batch transformations on SageMaker. Parsing texts with escape characters was not working properly.
We would like to thank all contributors to this release: