aboutsummaryrefslogtreecommitdiffstats
path: root/model-evaluation/src/main
Commit message (Collapse)AuthorAgeFilesLines
* Reapply "Bjorncs/embedder onnx gpu"Bjørn Christian Seime2023-05-091-1/+1
| | | | Updated model-integration to depend on configdefinitions instead of searchcore as OnnxModelsConfig is moved there.
* Revert "Bjorncs/embedder onnx gpu"Geir Storli2023-05-081-1/+1
|
* Require GPU when available for ONNX evaluation in global-phase and embeddersBjørn Christian Seime2023-05-081-1/+1
|
* check ONNX model input type with isAssignableToArne Juul2023-03-211-3/+5
|
* use withTransformedExpressions for wiringArne Juul2023-03-201-14/+5
|
* handle DynamicTensorArne Juul2023-03-171-0/+8
|
* * make InputRecorder handle slice with embedded expressionsArne Juul2023-03-162-1/+29
| | | | * special handling for Generate and Slice in BindingExtractor
* use OnnxExpressionNodeArne Juul2023-03-083-5/+75
|
* add ExpressionNode computing an output from an ONNX modelArne Juul2023-03-082-0/+121
|
* refactor binding extraction, with cachingArne Juul2023-03-083-118/+199
|
* pick up and remember declared returned types for functionsArne Juul2023-03-071-5/+9
|
* allow dollar in function namesArne Juul2023-03-031-4/+4
|
* use common utility and constantArne Juul2023-03-022-3/+10
|
* only optimize functions that have a contextPrototypeArne Juul2023-03-011-4/+8
|
* fix regexp patterns that only allowed exactly two argumentsArne Juul2023-03-011-4/+4
|
* Replace `OnnxEvaluatorCache` with OnnxRuntimeBjørn Christian Seime2023-02-273-21/+21
| | | | | | Require an `OnnxRuntime` instance to create `OnnxEvaluator` instances. Cache underlying `OrtSession` instead of `OnnxEvaluator`. Move static helpers for checking Onnx runtime availability from `OnnxEvaluator` to `OnnxRuntime`.
* Merge pull request #26188 from vespa-engine/arnej/pick-up-declare-feature-typesJon Bratseth2023-02-272-3/+30
|\ | | | | pick up declared type of query/attribute features,
| * pick up declared type of query/attribute features,Arne Juul2023-02-262-3/+30
| | | | | | | | and handle globalphase not wrapped in rankExpression().
* | replace assert -> check and throwArne Juul2023-02-261-2/+4
|/
* handle non-identifier onnx input/output names: instead of the conflictingArne Juul2023-02-221-16/+27
| | | | | ad-hoc code in OnnxEvaluator, do it as part of general input/output mapping in OnnxModel.
* add configurable input/output mappingsArne Juul2023-02-222-4/+122
|
* Cache Onnx model instancesBjørn Christian Seime2023-02-224-12/+33
| | | | | | Manage lifecycle of OnnxEvaluator instances explicitly to allow instances to be cached without use WeakHashmap/finalizers. Inject shared Onnx model cache in ModelsEvaluator.
* pick up specified name of output from onnx modelArne Juul2023-02-111-1/+7
|
* Merge pull request #25993 from ↵Henning Baldersheim2023-02-111-0/+12
|\ | | | | | | | | vespa-engine/arnej/get-global-evaluator-from-registry-2 handle that the component graph contains multiple ModelsEvaluator instances
| * handle that the component graph contains multiple ModelsEvaluator instances ↵Arne Juul2023-02-101-0/+12
| | | | | | | | (take 2)
* | also handle JSON-serialization of constantsArne Juul2023-02-111-3/+6
|/
* Revert "handle that the component graph contains multiple ModelsEvaluator ↵Henning Baldersheim2023-02-101-10/+0
| | | | instances"
* handle that the component graph contains multiple ModelsEvaluator instancesArne Juul2023-02-101-0/+10
|
* Allow fallback to CPU if nodes are provisioned without GPUMartin Polden2023-02-081-1/+1
|
* Remove 'required' attributeMartin Polden2023-01-261-1/+1
|
* Support configuration of GPU device to use in ONNX modelMartin Polden2023-01-231-1/+1
|
* Support direct tensor renderingJon Bratseth2023-01-141-9/+9
|
* sort argumentsArne Juul2022-12-151-3/+9
|
* Use well defined order where we output text and generate config. Makes ↵Henning Baldersheim2022-12-014-12/+12
| | | | config stable and simple tests predictable.
* Revert "Revert "- Reduce usage of guava.""Henning Baldersheim2022-12-015-56/+45
|
* Revert "- Reduce usage of guava."Henning Baldersheim2022-12-015-45/+56
|
* - Reduce usage of guava.Henning Baldersheim2022-12-015-56/+45
| | | | - Ensure that tests relying on order are determinsitic.
* Reduce the simple usage of guava where java has caught upHenning Baldersheim2022-11-301-21/+8
|
* Use consistent terminologyJon Bratseth2022-09-291-3/+3
|
* Use tensor vocabularyJon Bratseth2022-09-281-10/+10
|
* Fold comparisons into the other operatorsJon Bratseth2022-09-281-21/+50
|
* Set short tensor format as defaultLester Solbakken2022-06-081-3/+3
|
* GC unused methodsHenning Baldersheim2022-06-081-7/+0
|
* onnxModel to onnx in summary/matchfeaturesLester Solbakken2022-06-081-1/+1
|
* Use '@Inject' from 'annotations' in multiple bundlesBjørn Christian Seime2022-05-061-1/+1
|
* Allow models referencing functions they don't defineJon Bratseth2022-04-291-1/+3
| | | | | | | | ML models may not be designed to be used standalone but only in context of another ranking expression, which then supplies functions referenced by the model. Tolerate such models in stateløess evaluation.
* Add custom `@Beta` annotationBjørn Christian Seime2021-12-032-2/+2
| | | | Replace use of Guava's `com.google.common.annotations.Beta` with custom annotation.
* Avoid using deprecated methodsHenning Baldersheim2021-11-031-2/+2
|
* Make StringValue symmetric so it can accept its output (toString) as input.Henning Baldersheim2021-11-031-1/+0
|
* Update Verizon Media copyright notices.gjoranv2021-10-071-1/+1
|