summaryrefslogtreecommitdiffstats
path: root/model-integration
Commit message (Collapse)AuthorAgeFilesLines
...
* Handle models requiring token type idsBjørn Christian Seime2023-05-112-13/+20
|
* Don't lower caseBjørn Christian Seime2023-05-111-1/+1
|
* Disable special tokens by defaultBjørn Christian Seime2023-05-111-0/+1
|
* Mark HF integration as betaBjørn Christian Seime2023-05-111-0/+2
|
* Make HF tokenizer a separate embedderBjørn Christian Seime2023-05-115-197/+6
|
* Don't specify both package and namespaceBjørn Christian Seime2023-05-112-1/+1
|
* Upgrade HF Tokenizer to 0.22.1Bjørn Christian Seime2023-05-081-1/+1
|
* Handle nullsBjørn Christian Seime2023-05-081-0/+4
|
* fixup! Require GPU when requested and available for Bert + HF embeddersBjørn Christian Seime2023-05-081-1/+1
|
* Require GPU when requested and available for Bert + HF embeddersBjørn Christian Seime2023-05-085-5/+6
|
* Require GPU when available for ONNX evaluation in global-phase and embeddersBjørn Christian Seime2023-05-083-5/+42
|
* Make thread pool size configurableBjørn Christian Seime2023-05-055-17/+24
|
* Make normalization optionalBjørn Christian Seime2023-05-052-2/+8
|
* Allow for manual configuration of GPUBjørn Christian Seime2023-05-052-1/+8
|
* Move config to same package as componentBjørn Christian Seime2023-05-052-1/+1
|
* Split out HF TokenizerBjørn Christian Seime2023-05-054-23/+174
|
* Put the openai client in a separate componentJon Bratseth2023-04-2515-482/+20
|
* Export APIJon Bratseth2023-04-204-1/+218
|
* Merge pull request #26777 from vespa-engine/bratseth/openai-clientJon Bratseth2023-04-1910-0/+312
|\ | | | | Llm completion abstraction and OpenAi implementation
| * Use record and use default record toStringJon Bratseth2023-04-191-9/+1
| |
| * Fix typoJon Bratseth2023-04-191-1/+1
| |
| * Avoid ignored directory nameJon Bratseth2023-04-192-2/+45
| |
| * Llm completion abstraction and OpenAi implementationJon Bratseth2023-04-199-0/+277
| |
* | Merge pull request #26753 from vespa-engine/bjorncs/global-phaseArne H Juul2023-04-191-5/+9
|\ \ | |/ |/| Use quarter vcpu by default if execution mode is parallel
| * Use quarter vcpu by default if execution mode is parallelBjørn Christian Seime2023-04-171-5/+9
| |
* | Merge pull request #26754 from vespa-engine/bratseth/jdk20Jon Bratseth2023-04-181-2/+2
|\ \ | | | | | | Build with jdk20
| * | Build with jdk20Jon Bratseth2023-04-171-2/+2
| | |
* | | Pull endtokenJon Bratseth2023-04-181-1/+1
| | |
* | | Revert "Merge pull request #26744 from ↵Jon Bratseth2023-04-182-8/+14
| |/ |/| | | | | | | | | | | vespa-engine/revert-26708-allow-start-end-sequence-tokens-as-args-bertbaseembedder" This reverts commit d025a93015e66efc0027d81a64e70530d6cb240e, reversing changes made to 4f2f29e1459b900d4b074f5cfc4c126837c54bfd.
* | Revert "Allow start end sequence tokens as args bertbaseembedder"Jon Bratseth2023-04-142-14/+8
|/
* Include createTokenTypeIdsconnell gough2023-04-131-0/+1
|
* Remove separator input and fix spelling errorconnell gough2023-04-131-3/+2
|
* Add default special tokensconnell gough2023-04-131-2/+2
|
* Add special tokens as arguments and allow tokenTypeIds to be nullconnell gough2023-04-132-7/+13
|
* Add lz4-java for xxhashBjørn Christian Seime2023-03-311-0/+5
|
* Support loading ONNX models through byte arrayBjørn Christian Seime2023-03-304-36/+162
| | | | Rewrite OnnxRuntimeTest to test through it's public API
* Don't reuse runtime between methodsBjørn Christian Seime2023-03-301-21/+18
| | | | Caching evaluators between test methods may have unwanted side effects
* Revert "Arnej/unify cell type conversion"Henning Baldersheim2023-03-123-44/+21
|
* cell-type conversions should matchArne Juul2023-03-102-20/+39
|
* add BFLOAT16 from newer onnx.proto versionArne Juul2023-03-101-1/+5
|
* Revert "Deopend on component to get AbstractComponent"Bjørn Christian Seime2023-02-281-6/+0
|
* Deopend on component to get AbstractComponentJon Bratseth2023-02-281-0/+6
|
* Extend AbstractComponent not AbstractResourceBjørn Christian Seime2023-02-283-9/+9
|
* Replace `OnnxEvaluatorCache` with OnnxRuntimeBjørn Christian Seime2023-02-2714-228/+325
| | | | | | 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`.
* handle non-identifier onnx input/output names: instead of the conflictingArne Juul2023-02-221-0/+30
| | | | | ad-hoc code in OnnxEvaluator, do it as part of general input/output mapping in OnnxModel.
* Cache Onnx model instancesBjørn Christian Seime2023-02-223-0/+141
| | | | | | Manage lifecycle of OnnxEvaluator instances explicitly to allow instances to be cached without use WeakHashmap/finalizers. Inject shared Onnx model cache in ModelsEvaluator.
* Make OnnxEvaluator closableBjørn Christian Seime2023-02-221-1/+12
|
* Implement equals()/hashCode()Bjørn Christian Seime2023-02-221-1/+17
| | | | Required for using instances as key in HashMap
* Add initial text generator componentLester Solbakken2023-02-189-2/+410
|
* ensure outputs with names as promised by getOutputInfo()Arne Juul2023-02-103-1/+123
|