diff options
Diffstat (limited to 'model-integration/src')
3 files changed, 2 insertions, 27 deletions
diff --git a/model-integration/src/main/java/ai/vespa/embedding/BertBaseEmbedder.java b/model-integration/src/main/java/ai/vespa/embedding/BertBaseEmbedder.java index 42e3d653359..1831903d626 100644 --- a/model-integration/src/main/java/ai/vespa/embedding/BertBaseEmbedder.java +++ b/model-integration/src/main/java/ai/vespa/embedding/BertBaseEmbedder.java @@ -2,6 +2,7 @@ package ai.vespa.embedding; import ai.vespa.modelintegration.evaluator.OnnxEvaluator; import ai.vespa.modelintegration.evaluator.OnnxEvaluatorOptions; +import com.yahoo.embedding.BertBaseEmbedderConfig; import com.yahoo.component.annotation.Inject; import com.yahoo.language.process.Embedder; import com.yahoo.language.wordpiece.WordPieceEmbedder; diff --git a/model-integration/src/main/resources/configdefinitions/bert-base-embedder.def b/model-integration/src/main/resources/configdefinitions/bert-base-embedder.def deleted file mode 100644 index 7e3ff151466..00000000000 --- a/model-integration/src/main/resources/configdefinitions/bert-base-embedder.def +++ /dev/null @@ -1,27 +0,0 @@ -package=ai.vespa.embedding - -# Transformer model settings -transformerModelUrl url - -# Max length of token sequence model can handle -transformerMaxTokens int default=384 - -# Pooling strategy -poolingStrategy enum { cls, mean } default=mean - -# Input names -transformerInputIds string default=input_ids -transformerAttentionMask string default=attention_mask -transformerTokenTypeIds string default=token_type_ids - -# Output name -transformerOutput string default=output_0 - -# Settings for ONNX model evaluation -onnxExecutionMode enum { parallel, sequential } default=sequential -onnxInterOpThreads int default=1 -onnxIntraOpThreads int default=-4 # n=number of threads -> n<0: CPUs/(-n), n==0: CPUs, n>0: n - -# Settings for wordpiece tokenizer -tokenizerVocabUrl url - diff --git a/model-integration/src/test/java/ai/vespa/embedding/BertBaseEmbedderTest.java b/model-integration/src/test/java/ai/vespa/embedding/BertBaseEmbedderTest.java index 0ecc78f7668..464e5941e89 100644 --- a/model-integration/src/test/java/ai/vespa/embedding/BertBaseEmbedderTest.java +++ b/model-integration/src/test/java/ai/vespa/embedding/BertBaseEmbedderTest.java @@ -2,6 +2,7 @@ package ai.vespa.embedding; import ai.vespa.modelintegration.evaluator.OnnxEvaluator; import com.yahoo.config.UrlReference; +import com.yahoo.embedding.BertBaseEmbedderConfig; import com.yahoo.tensor.Tensor; import com.yahoo.tensor.TensorType; import org.junit.Test; |