[BUG] Integration tests are failing with Embedding dimension error.

[ryanbogan]

What is the bug?

Integration tests are failing for the repository with the following error:

{"error":{"root_cause":[{"type":"illegal_argument_exception","reason":"embedding dimension is null"}],"type":"illegal_argument_exception","reason":"embedding dimension is null"},"status":400}

How can one reproduce the bug?

./gradlew integTest

What is the expected behavior?

Integration tests should pass

Reference: https://github.com/opensearch-project/neural-search/actions/runs/15150232212/job/42660197102?pr=1336

#1336

[junqiu-lei]

@ryanbogan With your code change in #1336 at my local, no issue found in the integTest:

(base) ➜  neural-search git:(upgrade_lucene) ✗ ./gradlew integTest
=======================================
OpenSearch Build Hamster says Hello!
  Gradle Version        : 8.10.2
  OS Info               : Mac OS X 15.4.1 (aarch64)
  JDK Version           : 21 (OpenJDK)
  JAVA_HOME             : /Users/junqiu/.sdkman/candidates/java/21.0.2-open
  Random Testing Seed   : 899F0D9EA6F94376
  In FIPS 140 mode      : false
=======================================

Deprecated Gradle features were used in this build, making it incompatible with Gradle 9.0.

You can use '--warning-mode all' to show the individual deprecation warnings and determine if they come from your own scripts or plugins.

For more on this, please refer to https://docs.gradle.org/8.10.2/userguide/command_line_interface.html#sec:command_line_warnings in the Gradle documentation.

BUILD SUCCESSFUL in 11m 30s
37 actionable tasks: 3 executed, 34 up-to-date

Might be some issue related Github CI resource overloaded after the Lucene change?

[ryanbogan]

@junqiu-lei That's interesting, when I run it locally it fails with the same error as the CI.

[junqiu-lei]

@ryanbogan With your code change in #1336 at my local, no issue found in the integTest:

(base) ➜  neural-search git:(upgrade_lucene) ✗ ./gradlew integTest
=======================================
OpenSearch Build Hamster says Hello!
  Gradle Version        : 8.10.2
  OS Info               : Mac OS X 15.4.1 (aarch64)
  JDK Version           : 21 (OpenJDK)
  JAVA_HOME             : /Users/junqiu/.sdkman/candidates/java/21.0.2-open
  Random Testing Seed   : 899F0D9EA6F94376
  In FIPS 140 mode      : false
=======================================

Deprecated Gradle features were used in this build, making it incompatible with Gradle 9.0.

You can use '--warning-mode all' to show the individual deprecation warnings and determine if they come from your own scripts or plugins.

For more on this, please refer to https://docs.gradle.org/8.10.2/userguide/command_line_interface.html#sec:command_line_warnings in the Gradle documentation.

BUILD SUCCESSFUL in 11m 30s
37 actionable tasks: 3 executed, 34 up-to-date

Might be some issue related Github CI resource overloaded after the Lucene change?

It should be failed... While I pulled and ran the test to my local with #1336, I forgot to revert back gradle.build changes which uses my local ml-commons artifact which commit is at c3b1fd6dfaed8a4745b63ac9b2e12a01947e1cff from last week.

I verified run the neural plugin integ test the with ml-commons' head commit dad243ff386484a07a237346b5a38efed784c891 from opensearch-project/ml-commons#3786 failed with same error, but the second head commit ade09e65639aaa83e68fdc2db00f04c35c0a5e19 can pass all neural integration test.

cc: opensearch-project/ml-commons#3786 owner @nathaliellenaa

[nathaliellenaa]

@ryanbogan @junqiu-lei In the UploadModelRequestBody.json, we should specify the function_name as TEXT_EMBEDDING. Due to the changes made in this PR, we check for function_name during model registration, and if we don't specify any, it will go the BaseModelConfig, which doesn't parse embedding_dimension. Here's the example of the model request body:

{
  "name": "traced_small_model",
  "version": "1.0.0",
  "model_format": "TORCH_SCRIPT",
  "function_name": "TEXT_EMBEDDING",
  "model_task_type": "text_embedding",
  "model_content_hash_value": "e13b74006290a9d0f58c1376f9629d4ebc05a0f9385f40db837452b167ae9021",
  "model_group_id": "%s",
  "model_config": {
    "model_type": "bert",
    "embedding_dimension": 768,
    "framework_type": "sentence_transformers",
    "all_config": "{\"architectures\":[\"BertModel\"],\"max_position_embeddings\":512,\"model_type\":\"bert\",\"num_attention_heads\":12,\"num_hidden_layers\":6}"
  },
  "url": "https://github.com/opensearch-project/ml-commons/blob/2.x/ml-algorithms/src/test/resources/org/opensearch/ml/engine/algorithms/text_embedding/traced_small_model.zip?raw=true"
}

I believe this should solve the issue. Let me know if you have any other questions/issue.

[ryanbogan]

@nathaliellenaa Thanks for the insight! I updated the file for the integ tests and bwc tests and so far it looks like tests are passing.

[nathaliellenaa]

Awesome!