diff options
Diffstat (limited to 'config-model/src/test/cfg/application/embed')
3 files changed, 44 insertions, 38 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 deleted file mode 100644 index 144dfbd0001..00000000000 --- a/config-model/src/test/cfg/application/embed/configdefinitions/embedding.bert-base-embedder.def +++ /dev/null @@ -1,30 +0,0 @@ -# Copy of this Vespa config stored here because Vespa config definitions are not -# available in unit tests, and are needed (by DomConfigPayloadBuilder.parseLeaf) -# Alternatively, 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/configdefinitions/sentence-embedder.def b/config-model/src/test/cfg/application/embed/configdefinitions/sentence-embedder.def new file mode 100644 index 00000000000..87b80f1051a --- /dev/null +++ b/config-model/src/test/cfg/application/embed/configdefinitions/sentence-embedder.def @@ -0,0 +1,26 @@ +package=ai.vespa.example.paragraph + +# WordPiece tokenizer vocabulary +vocab model + +model model + +myValue string + +# Max length of token sequence model can handle +transforerMaxTokens int default=128 + +# Pooling strategy +poolingStrategy enum { cls, mean } default=mean + +# Input names +transformerInputIds string default=input_ids +transformerAttentionMask string default=attention_mask + +# Output name +transformerOutput string default=last_hidden_state + +# Settings for ONNX model evaluation +onnxExecutionMode enum { parallel, sequential } default=sequential +onnxInterOpThreads int default=1 +onnxIntraOpThreads int default=-4 diff --git a/config-model/src/test/cfg/application/embed/services.xml b/config-model/src/test/cfg/application/embed/services.xml index 99c89bc4324..6823ef900ae 100644 --- a/config-model/src/test/cfg/application/embed/services.xml +++ b/config-model/src/test/cfg/application/embed/services.xml @@ -16,6 +16,7 @@ <onnx-intraop-threads>10</onnx-intraop-threads> <onnx-interop-threads>8</onnx-interop-threads> <onnx-gpu-device>1</onnx-gpu-device> + <pooling-strategy>mean</pooling-strategy> </component> <component id="hf-tokenizer" type="hugging-face-tokenizer"> @@ -25,15 +26,24 @@ <truncation>true</truncation> </component> - <component id="transformer" class="ai.vespa.embedding.BertBaseEmbedder" bundle="model-integration"> - <config name="embedding.bert-base-embedder"> - <!-- model specifics --> - <transformerModel model-id="minilm-l6-v2" url="application-url"/> - <tokenizerVocab path="files/vocab.txt"/> + <component id="bert-embedder" type="bert-embedder"> + <!-- model specifics --> + <transformer-model model-id="minilm-l6-v2" url="application-url"/> + <tokenizer-vocab path="files/vocab.txt"/> + <max-tokens>512</max-tokens> + <transformer-input-ids>my_input_ids</transformer-input-ids> + <transformer-attention-mask>my_attention_mask</transformer-attention-mask> + <transformer-token-type-ids>my_token_type_ids</transformer-token-type-ids> + <transformer-output>my_output</transformer-output> + <transformer-start-sequence-token>101</transformer-start-sequence-token> + <transformer-end-sequence-token>102</transformer-end-sequence-token> - <!-- tunable parameters: number of threads etc --> - <onnxIntraOpThreads>4</onnxIntraOpThreads> - </config> + + <!-- tunable parameters: number of threads etc --> + <onnx-execution-mode>parallel</onnx-execution-mode> + <onnx-intraop-threads>4</onnx-intraop-threads> + <onnx-interop-threads>8</onnx-interop-threads> + <onnx-gpu-device>1</onnx-gpu-device> </component> <nodes> |