Remove/rewrite unit tests with old mechanism for configuring Bert embedder

6 files changed, 57 insertions, 79 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 0ce61b8ddf8..6823ef900ae 100644
--- a/config-model/src/test/cfg/application/embed/services.xml
+++ b/config-model/src/test/cfg/application/embed/services.xml
@@ -46,17 +46,6 @@
       <onnx-gpu-device>1</onnx-gpu-device>
     </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"/>
-
-        <!-- tunable parameters: number of threads etc -->
-        <onnxIntraOpThreads>4</onnxIntraOpThreads>
-      </config>
-    </component>
-
     <nodes>
       <node hostalias="node1" />
     </nodes>
diff --git a/config-model/src/test/cfg/application/embed_cloud_only/configdefinitions/embedding.bert-base-embedder.def b/config-model/src/test/cfg/application/embed_cloud_only/configdefinitions/embedding.bert-base-embedder.def
deleted file mode 100644
index 144dfbd0001..00000000000
--- a/config-model/src/test/cfg/application/embed_cloud_only/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_cloud_only/configdefinitions/sentence-embedder.def b/config-model/src/test/cfg/application/embed_cloud_only/configdefinitions/sentence-embedder.def
new file mode 100644
index 00000000000..87b80f1051a
--- /dev/null
+++ b/config-model/src/test/cfg/application/embed_cloud_only/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_cloud_only/services.xml b/config-model/src/test/cfg/application/embed_cloud_only/services.xml
index 57db4f5bfae..e203ec56669 100644
--- a/config-model/src/test/cfg/application/embed_cloud_only/services.xml
+++ b/config-model/src/test/cfg/application/embed_cloud_only/services.xml
@@ -4,14 +4,11 @@
 
   <container version="1.0">
 
-    <component id="transformer" class="ai.vespa.embedding.BertBaseEmbedder" bundle="model-integration">
-      <config name="embedding.bert-base-embedder">
-        <!-- No fallback to url or path when deploying outside cloud -->
-        <transformerModel model-id="minilm-l6-v2"/>
-        <tokenizerVocab path="files/vocab.txt"/>
-
-        <!-- tunable parameters: number of threads etc -->
-        <onnxIntraOpThreads>4</onnxIntraOpThreads>
+    <component id="transformer" class="ai.vespa.example.paragraph.ApplicationSpecificEmbedder" bundle="app">
+      <config name='ai.vespa.example.paragraph.sentence-embedder'>
+        <model model-id="minilm-l6-v2"/>
+        <vocab path="files/vocab.txt"/>
+        <myValue>foo</myValue>
       </config>
     </component>