Merge branch 'master' into olaa/dataplane-proxy-config

8 files changed, 104 insertions, 1 deletions

diff --git a/configdefinitions/pom.xml b/configdefinitions/pom.xml
index 51221c4899f..e163f0292b5 100644
--- a/configdefinitions/pom.xml
+++ b/configdefinitions/pom.xml
@@ -32,6 +32,10 @@
         <groupId>com.yahoo.vespa</groupId>
         <artifactId>bundle-plugin</artifactId>
         <extensions>true</extensions>
+        <configuration>
+          <bundleType>CORE</bundleType>
+          <suppressWarningMissingImportPackages>true</suppressWarningMissingImportPackages>
+        </configuration>
       </plugin>
       <plugin>
         <groupId>org.apache.maven.plugins</groupId>
diff --git a/configdefinitions/src/main/java/com/yahoo/embedding/huggingface/package-info.java b/configdefinitions/src/main/java/com/yahoo/embedding/huggingface/package-info.java
new file mode 100644
index 00000000000..7bcc994e616
--- /dev/null
+++ b/configdefinitions/src/main/java/com/yahoo/embedding/huggingface/package-info.java
@@ -0,0 +1,9 @@
+// Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
+
+/**
+ * @author bjorncs
+ */
+@ExportPackage
+package com.yahoo.embedding.huggingface;
+
+import com.yahoo.osgi.annotation.ExportPackage;
\ No newline at end of file
diff --git a/configdefinitions/src/main/java/com/yahoo/language/huggingface/config/package-info.java b/configdefinitions/src/main/java/com/yahoo/language/huggingface/config/package-info.java
new file mode 100644
index 00000000000..fb9048b5fb4
--- /dev/null
+++ b/configdefinitions/src/main/java/com/yahoo/language/huggingface/config/package-info.java
@@ -0,0 +1,9 @@
+// Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
+
+/**
+ * @author bjorncs
+ */
+@ExportPackage
+package com.yahoo.language.huggingface.config;
+
+import com.yahoo.osgi.annotation.ExportPackage;
\ No newline at end of file
diff --git a/configdefinitions/src/vespa/CMakeLists.txt b/configdefinitions/src/vespa/CMakeLists.txt
index 6a737e6f57d..496e92916ee 100644
--- a/configdefinitions/src/vespa/CMakeLists.txt
+++ b/configdefinitions/src/vespa/CMakeLists.txt
@@ -84,4 +84,8 @@ vespa_generate_config(configdefinitions proton.def)
 install_config_definition(proton.def vespa.config.search.core.proton.def)
 vespa_generate_config(configdefinitions hwinfo.def)
 vespa_generate_config(configdefinitions dataplane-proxy.def)
-install_config_definition(dataplane-proxy.def cloud.config.dataplane-proxy.def)
\ No newline at end of file
+install_config_definition(dataplane-proxy.def cloud.config.dataplane-proxy.def)
+install_config_definition(hugging-face-embedder.def embedding.huggingface.hugging-face-embedder.def)
+install_config_definition(hugging-face-tokenizer.def language.huggingface.config.hugging-face-tokenizer.def)
+install_config_definition(bert-base-embedder.def embedding.bert-base-embedder.def)
+
diff --git a/configdefinitions/src/vespa/bert-base-embedder.def b/configdefinitions/src/vespa/bert-base-embedder.def
new file mode 100644
index 00000000000..2d8e840377b
--- /dev/null
+++ b/configdefinitions/src/vespa/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
diff --git a/configdefinitions/src/vespa/fleetcontroller.def b/configdefinitions/src/vespa/fleetcontroller.def
index 93a20e4ee0d..c3e161eb038 100644
--- a/configdefinitions/src/vespa/fleetcontroller.def
+++ b/configdefinitions/src/vespa/fleetcontroller.def
@@ -29,6 +29,7 @@ master_zookeeper_cooldown_period double default=60.0
 ## If set to 1, only master will gather state. If set higher, others will
 ## also do so, prioritizing those fleetcontrollers likely to be the ones to
 ## take over if the master fails.
+# TODO: Deprecated, not used anymore, remove in Vespa 9
 state_gather_count int default=1
 
 ## Location of ZooKeeper servers
diff --git a/configdefinitions/src/vespa/hugging-face-embedder.def b/configdefinitions/src/vespa/hugging-face-embedder.def
new file mode 100644
index 00000000000..7ea4227b3cd
--- /dev/null
+++ b/configdefinitions/src/vespa/hugging-face-embedder.def
@@ -0,0 +1,31 @@
+namespace=embedding.huggingface
+
+# Path to tokenizer.json
+tokenizerPath model
+
+# Path to model.onnx
+transformerModel  model
+
+# Max length of token sequence model can handle
+transformerMaxTokens int default=512
+
+# Input names
+transformerInputIds      string default=input_ids
+transformerAttentionMask string default=attention_mask
+transformerTokenTypeIds  string default=token_type_ids
+
+# Output name
+transformerOutput string default=last_hidden_state
+
+
+# Normalize tensors from tokenizer
+normalize bool default=false
+
+poolingStrategy enum { cls, mean } default=mean
+
+# Settings for ONNX model evaluation
+transformerExecutionMode enum { parallel, sequential } default=sequential
+transformerInterOpThreads int default=1
+transformerIntraOpThreads int default=-4
+# GPU device id, -1 for CPU
+transformerGpuDevice      int default=0
diff --git a/configdefinitions/src/vespa/hugging-face-tokenizer.def b/configdefinitions/src/vespa/hugging-face-tokenizer.def
new file mode 100644
index 00000000000..18b3631e494
--- /dev/null
+++ b/configdefinitions/src/vespa/hugging-face-tokenizer.def
@@ -0,0 +1,13 @@
+# Copyright Yahoo. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
+
+namespace=language.huggingface.config
+
+# The language a model is for, one of the language tags in com.yahoo.language.Language.
+# Use "unknown" for models to be used with any language.
+model[].language string
+# The path to the model relative to the application package root
+model[].path model
+
+addSpecialTokens bool default=true
+maxLength int default=-1
+truncation bool default=false
\ No newline at end of file