Skip to content

Commit b7d10b8

Browse files
committed
Pass through chunking settings and provide default for default endpoint
1 parent 27ca440 commit b7d10b8

File tree

3 files changed

+15
-9
lines changed

3 files changed

+15
-9
lines changed

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/elastic/ElasticInferenceService.java

Lines changed: 4 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -201,7 +201,8 @@ private static Map<String, DefaultModelConfig> initDefaultEndpoints(
201201
),
202202
EmptyTaskSettings.INSTANCE,
203203
EmptySecretSettings.INSTANCE,
204-
elasticInferenceServiceComponents
204+
elasticInferenceServiceComponents,
205+
ChunkingSettingsBuilder.DEFAULT_SETTINGS
205206
),
206207
MinimalServiceSettings.textEmbedding(
207208
NAME,
@@ -482,7 +483,8 @@ private static ElasticInferenceServiceModel createModel(
482483
taskSettings,
483484
secretSettings,
484485
elasticInferenceServiceComponents,
485-
context
486+
context,
487+
chunkingSettings
486488
);
487489
default -> throw new ElasticsearchStatusException(failureMessage, RestStatus.BAD_REQUEST);
488490
};

x-pack/plugin/inference/src/main/java/org/elasticsearch/xpack/inference/services/elastic/densetextembeddings/ElasticInferenceServiceDenseTextEmbeddingsModel.java

Lines changed: 8 additions & 6 deletions
Original file line numberDiff line numberDiff line change
@@ -9,6 +9,7 @@
99

1010
import org.elasticsearch.ElasticsearchStatusException;
1111
import org.elasticsearch.core.Nullable;
12+
import org.elasticsearch.inference.ChunkingSettings;
1213
import org.elasticsearch.inference.EmptySecretSettings;
1314
import org.elasticsearch.inference.EmptyTaskSettings;
1415
import org.elasticsearch.inference.ModelConfigurations;
@@ -39,17 +40,18 @@ public ElasticInferenceServiceDenseTextEmbeddingsModel(
3940
Map<String, Object> taskSettings,
4041
Map<String, Object> secrets,
4142
ElasticInferenceServiceComponents elasticInferenceServiceComponents,
42-
ConfigurationParseContext context
43+
ConfigurationParseContext context,
44+
ChunkingSettings chunkingSettings
4345
) {
4446
this(
4547
inferenceEntityId,
4648
taskType,
4749
service,
4850
ElasticInferenceServiceDenseTextEmbeddingsServiceSettings.fromMap(serviceSettings, context),
49-
// TODO: we probably want dense embeddings task settings
5051
EmptyTaskSettings.INSTANCE,
5152
EmptySecretSettings.INSTANCE,
52-
elasticInferenceServiceComponents
53+
elasticInferenceServiceComponents,
54+
chunkingSettings
5355
);
5456
}
5557

@@ -58,13 +60,13 @@ public ElasticInferenceServiceDenseTextEmbeddingsModel(
5860
TaskType taskType,
5961
String service,
6062
ElasticInferenceServiceDenseTextEmbeddingsServiceSettings serviceSettings,
61-
// TODO: we probably want dense embeddings task settings
6263
@Nullable TaskSettings taskSettings,
6364
@Nullable SecretSettings secretSettings,
64-
ElasticInferenceServiceComponents elasticInferenceServiceComponents
65+
ElasticInferenceServiceComponents elasticInferenceServiceComponents,
66+
ChunkingSettings chunkingSettings
6567
) {
6668
super(
67-
new ModelConfigurations(inferenceEntityId, taskType, service, serviceSettings, taskSettings),
69+
new ModelConfigurations(inferenceEntityId, taskType, service, serviceSettings, taskSettings, chunkingSettings),
6870
new ModelSecrets(secretSettings),
6971
serviceSettings,
7072
elasticInferenceServiceComponents

x-pack/plugin/inference/src/test/java/org/elasticsearch/xpack/inference/services/elastic/densetextembeddings/ElasticInferenceServiceDenseTextEmbeddingsModelTests.java

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -11,6 +11,7 @@
1111
import org.elasticsearch.inference.EmptyTaskSettings;
1212
import org.elasticsearch.inference.SimilarityMeasure;
1313
import org.elasticsearch.inference.TaskType;
14+
import org.elasticsearch.xpack.inference.chunking.ChunkingSettingsBuilder;
1415
import org.elasticsearch.xpack.inference.services.elastic.ElasticInferenceServiceComponents;
1516
import org.elasticsearch.xpack.inference.services.settings.RateLimitSettings;
1617

@@ -30,7 +31,8 @@ public static ElasticInferenceServiceDenseTextEmbeddingsModel createModel(String
3031
),
3132
EmptyTaskSettings.INSTANCE,
3233
EmptySecretSettings.INSTANCE,
33-
ElasticInferenceServiceComponents.of(url)
34+
ElasticInferenceServiceComponents.of(url),
35+
ChunkingSettingsBuilder.DEFAULT_SETTINGS
3436
);
3537
}
3638

0 commit comments

Comments
 (0)