diff options
author | Bjørn Christian Seime <bjorncs@yahooinc.com> | 2023-06-05 15:31:33 +0200 |
---|---|---|
committer | Bjørn Christian Seime <bjorncs@yahooinc.com> | 2023-06-05 15:31:33 +0200 |
commit | f944b96338725a0a75bbe52922f98f9342abcdd4 (patch) | |
tree | 433ff453847c9f94ca18d9c9c44f778bcdbb9cda /configdefinitions | |
parent | 8226d4766007f0091826044c126655f388522a43 (diff) |
Move config definition to `configdefinitions`
Diffstat (limited to 'configdefinitions')
-rw-r--r-- | configdefinitions/src/vespa/embedding.bert-base-embedder.def | 32 |
1 files changed, 32 insertions, 0 deletions
diff --git a/configdefinitions/src/vespa/embedding.bert-base-embedder.def b/configdefinitions/src/vespa/embedding.bert-base-embedder.def new file mode 100644 index 00000000000..2d8e840377b --- /dev/null +++ b/configdefinitions/src/vespa/embedding.bert-base-embedder.def @@ -0,0 +1,32 @@ + +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 + +# special token ids +transformerStartSequenceToken int default=101 +transformerEndSequenceToken int default=102 + +# 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 +# GPU device id, -1 for CPU +onnxGpuDevice int default=0 |