From c172d09ebb17b47ebb2e75bb04d09a8533350450 Mon Sep 17 00:00:00 2001
From: Jon Bratseth <bratseth@gmail.com>
Date: Wed, 14 Sep 2022 17:51:12 +0200
Subject: Move the BERT config def to model-integration

---
 .../src/vespa/embedding.bert-base-embedder.def     | 27 ----------------------
 model-integration/CMakeLists.txt                   |  4 +++-
 .../java/ai/vespa/embedding/BertBaseEmbedder.java  |  4 ++--
 .../embedding.bert-base-embedder.def               | 27 ++++++++++++++++++++++
 4 files changed, 32 insertions(+), 30 deletions(-)
 delete mode 100644 configdefinitions/src/vespa/embedding.bert-base-embedder.def
 create mode 100644 model-integration/src/main/resources/configdefinitions/embedding.bert-base-embedder.def

diff --git a/configdefinitions/src/vespa/embedding.bert-base-embedder.def b/configdefinitions/src/vespa/embedding.bert-base-embedder.def
deleted file mode 100644
index 14d953eeef9..00000000000
--- a/configdefinitions/src/vespa/embedding.bert-base-embedder.def
+++ /dev/null
@@ -1,27 +0,0 @@
-
-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/model-integration/CMakeLists.txt b/model-integration/CMakeLists.txt
index 4423746ccbc..9a19c5c31b0 100644
--- a/model-integration/CMakeLists.txt
+++ b/model-integration/CMakeLists.txt
@@ -3,4 +3,6 @@ install_jar(model-integration-jar-with-dependencies.jar)
 
 vespa_install_script(src/main/python/vespa-convert-tf2onnx.py vespa-convert-tf2onnx bin)
 
-install(FILES src/main/config/model-integration.xml DESTINATION conf/configserver-app)
\ No newline at end of file
+install(FILES src/main/config/model-integration.xml DESTINATION conf/configserver-app)
+
+install_config_definitions()
\ No newline at end of file
diff --git a/model-integration/src/main/java/ai/vespa/embedding/BertBaseEmbedder.java b/model-integration/src/main/java/ai/vespa/embedding/BertBaseEmbedder.java
index 149598ee2dd..4a1f59cc5cf 100644
--- a/model-integration/src/main/java/ai/vespa/embedding/BertBaseEmbedder.java
+++ b/model-integration/src/main/java/ai/vespa/embedding/BertBaseEmbedder.java
@@ -17,7 +17,7 @@ import java.util.Map;
 
 /**
  * A BERT Base compatible embedder. This embedder uses a WordPiece embedder to
- * produce a token sequence that is input to a transformer model. A BERT base
+ * produce a token sequence that is then input to a transformer model. A BERT base
  * compatible transformer model must have three inputs:
  *
  *  - A token sequence (input_ids)
@@ -76,7 +76,7 @@ public class BertBaseEmbedder implements Embedder {
     private void validateName(Map<String, TensorType> types, String name, String type) {
         if ( ! types.containsKey(name)) {
             throw new IllegalArgumentException("Model does not contain required " + type + ": '" + name + "'. " +
-                    "Model contains: " + String.join(",", types.keySet()));
+                                               "Model contains: " + String.join(",", types.keySet()));
         }
     }
 
diff --git a/model-integration/src/main/resources/configdefinitions/embedding.bert-base-embedder.def b/model-integration/src/main/resources/configdefinitions/embedding.bert-base-embedder.def
new file mode 100644
index 00000000000..14d953eeef9
--- /dev/null
+++ b/model-integration/src/main/resources/configdefinitions/embedding.bert-base-embedder.def
@@ -0,0 +1,27 @@
+
+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
+
-- 
cgit v1.2.3