diff options
Diffstat (limited to 'config-model/src/test/cfg')
4 files changed, 35 insertions, 8 deletions
diff --git a/config-model/src/test/cfg/application/embed/configdefinitions/embedding.bert-base-embedder.def b/config-model/src/test/cfg/application/embed/configdefinitions/embedding.bert-base-embedder.def new file mode 100644 index 00000000000..a6544187140 --- /dev/null +++ b/config-model/src/test/cfg/application/embed/configdefinitions/embedding.bert-base-embedder.def @@ -0,0 +1,30 @@ +# Copy of this Vespa config stored here because Vespa config definitions are not +# available in unit tests, and are needed (by DomConfigPayloadBuilder.parseLeaf) +# Alternativ ely, we could make that not need it as it is not strictly necessaery. + +namespace=embedding + +# Wordpiece tokenizer +tokenizerVocab model + +transformerModel model + +# 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 + diff --git a/config-model/src/test/cfg/application/embed/services.xml b/config-model/src/test/cfg/application/embed/services.xml index 88558ace4bf..cdbcfd67f02 100644 --- a/config-model/src/test/cfg/application/embed/services.xml +++ b/config-model/src/test/cfg/application/embed/services.xml @@ -7,7 +7,7 @@ <component id="transformer" class="ai.vespa.embedding.BertBaseEmbedder" bindle="model-integration"> <config name="embedding.bert-base-embedder"> <!-- model specifics --> - <transformerModel id="minilm-l6-v2" url="application-url"/> + <transformerModel model-id="minilm-l6-v2" url="application-url"/> <tokenizerVocab path="files/vocab.txt"/> <!-- tunable parameters: number of threads etc --> diff --git a/config-model/src/test/cfg/application/embed_generic/configdefinitions/sentence-embedder.def b/config-model/src/test/cfg/application/embed_generic/configdefinitions/sentence-embedder.def index 81fc88dbf01..87b80f1051a 100644 --- a/config-model/src/test/cfg/application/embed_generic/configdefinitions/sentence-embedder.def +++ b/config-model/src/test/cfg/application/embed_generic/configdefinitions/sentence-embedder.def @@ -1,12 +1,9 @@ package=ai.vespa.example.paragraph -# Settings for wordpiece tokenizer -vocabPath path -vocabUrl string +# WordPiece tokenizer vocabulary +vocab model -# Transformer model settings -modelPath path -modelUrl string +model model myValue string diff --git a/config-model/src/test/cfg/application/embed_generic/services.xml b/config-model/src/test/cfg/application/embed_generic/services.xml index ea430f24e2f..d2c22c03343 100644 --- a/config-model/src/test/cfg/application/embed_generic/services.xml +++ b/config-model/src/test/cfg/application/embed_generic/services.xml @@ -8,7 +8,7 @@ class='ai.vespa.example.paragraph.ApplicationSpecificEmbedder' bundle='exampleEmbedder'> <config name='ai.vespa.example.paragraph.sentence-embedder'> - <model id="minilm-l6-v2" url="application-url" /> + <model model-id="minilm-l6-v2" url="application-url" /> <vocab path="files/vocab.txt"/> <myValue>foo</myValue> </config> |