[ML] _infer using XML-RoBERTa model failed when input contains emojis

[wwang500]

Elasticsearch Version

8.12.0

OS Version

Linux x86

Problem Description

While running inference using XML-RoBERTa model: 02shanky/finetuned-twitter-xlm-roberta-base-emotion, if input contains certain mixed emojis and strings , such as: 😱😱😱😱😱😱this is weird, inference will fail.

The error message is: [array_index_out_of_bounds_exception Root causes: array_index_out_of_bounds_exception: Index 33 out of bounds for length 32]: Index 33 out of bounds for length 32

Steps to Reproduce

1. Import 02shanky/finetuned-twitter-xlm-roberta-base-emotion model from huggingface: eland_import_hub_model --url {es_url} -u {es_user} -p {es_password} --insecure --hub-model-id 02shanky/finetuned-twitter-xlm-roberta-base-emotion --task-type text_classification --start
2. Go to Machine Learning -> Model Management -> Trained Models: test the model using the text: "😱😱😱😱😱😱this is weird"

Observed

Error log

[2024-01-31T10:38:25,435][WARN ][r.suppressed             ] [node-0] path: /_ml/trained_models/02shanky__finetuned-twitter-xlm-roberta-base-emotion/_infer, params: {model_id=02shanky__finetuned-twitter-xlm-roberta-base-emotion}, status: 500
java.lang.ArrayIndexOutOfBoundsException: Index 33 out of bounds for length 32
        at org.elasticsearch.xpack.ml.inference.nlp.tokenizers.UnigramTokenizer.tokenize(UnigramTokenizer.java:283) ~[?:?]
        at org.elasticsearch.xpack.ml.inference.nlp.tokenizers.UnigramTokenizer.incrementToken(UnigramTokenizer.java:223) ~[?:?]
        at org.elasticsearch.xpack.ml.inference.nlp.tokenizers.XLMRobertaTokenizer.innerTokenize(XLMRobertaTokenizer.java:173) ~[?:?]
        at org.elasticsearch.xpack.ml.inference.nlp.tokenizers.NlpTokenizer.tokenize(NlpTokenizer.java:60) ~[?:?]
        at org.elasticsearch.xpack.ml.inference.nlp.tokenizers.XLMRobertaTokenizer.lambda$requestBuilder$0(XLMRobertaTokenizer.java:132) ~[?:?]
        at java.util.stream.ReferencePipeline$7$1.accept(ReferencePipeline.java:273) ~[?:?]
        at java.util.stream.IntPipeline$1$1.accept(IntPipeline.java:180) ~[?:?]
        at java.util.stream.Streams$RangeIntSpliterator.forEachRemaining(Streams.java:104) ~[?:?]
        at java.util.Spliterator$OfInt.forEachRemaining(Spliterator.java:712) ~[?:?]
        at java.util.stream.AbstractPipeline.copyInto(AbstractPipeline.java:509) ~[?:?]
        at java.util.stream.AbstractPipeline.wrapAndCopyInto(AbstractPipeline.java:499) ~[?:?]
        at java.util.stream.ReduceOps$ReduceOp.evaluateSequential(ReduceOps.java:921) ~[?:?]
        at java.util.stream.AbstractPipeline.evaluate(AbstractPipeline.java:234) ~[?:?]
        at java.util.stream.ReferencePipeline.collect(ReferencePipeline.java:682) ~[?:?]
        at org.elasticsearch.xpack.ml.inference.nlp.tokenizers.XLMRobertaTokenizer.lambda$requestBuilder$1(XLMRobertaTokenizer.java:133) ~[?:?]
        at org.elasticsearch.xpack.ml.inference.deployment.InferencePyTorchAction.doRun(InferencePyTorchAction.java:122) ~[?:?]
        at org.elasticsearch.common.util.concurrent.ThreadContext$ContextPreservingAbstractRunnable.doRun(ThreadContext.java:984) ~[elasticsearch-8.13.0-SNAPSHOT.jar:?]
        at org.elasticsearch.common.util.concurrent.AbstractRunnable.run(AbstractRunnable.java:26) ~[elasticsearch-8.13.0-SNAPSHOT.jar:?]
        at org.elasticsearch.xpack.ml.inference.pytorch.PriorityProcessWorkerExecutorService$OrderedRunnable.run(PriorityProcessWorkerExecutorService.java:54) ~[?:?]
        at org.elasticsearch.xpack.ml.job.process.AbstractProcessWorkerExecutorService.start(AbstractProcessWorkerExecutorService.java:111) ~[?:?]
        at java.util.concurrent.Executors$RunnableAdapter.call(Executors.java:572) ~[?:?]
        at java.util.concurrent.FutureTask.run(FutureTask.java:317) ~[?:?]
        at org.elasticsearch.common.util.concurrent.ThreadContext$ContextPreservingRunnable.run(ThreadContext.java:917) ~[elasticsearch-8.13.0-SNAPSHOT.jar:?]
        at java.util.concurrent.ThreadPoolExecutor.runWorker(ThreadPoolExecutor.java:1144) ~[?:?]
        at java.util.concurrent.ThreadPoolExecutor$Worker.run(ThreadPoolExecutor.java:642) ~[?:?]
        at java.lang.Thread.run(Thread.java:1583) ~[?:?]

[elasticsearchmachine]

Pinging @elastic/ml-core (Team:ML)

[joeafari]

Hey team,

This issue can occur with only emojis as well

I'd like to clarify that issue also occurs when there're only emojis

Also, different models lead to the same issue

• https://huggingface.co/ml6team/xlm-roberta-base-nl-emoji-ner
• https://huggingface.co/MilaNLProc/xlm-emo-t
• https://huggingface.co/02shanky/finetuned-twitter-xlm-roberta-base-emotion