From adcb1d4d55e71d78c662f798b033d3abea0d4b9e Mon Sep 17 00:00:00 2001
From: Jon Bratseth <bratseth@gmail.com>
Date: Wed, 31 Aug 2022 22:50:14 +0200
Subject: Add 'model' config type

---
 .../embedding.bert-base-embedder.def               | 30 ++++++++++++++++++++++
 .../src/test/cfg/application/embed/services.xml    |  2 +-
 2 files changed, 31 insertions(+), 1 deletion(-)
 create mode 100644 config-model/src/test/cfg/application/embed/configdefinitions/embedding.bert-base-embedder.def

(limited to 'config-model/src/test/cfg/application/embed')

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 -->
-- 
cgit v1.2.3