SPARKNLP-1000: Fix No Operation named [init_all_tables] for GPT2 #14177
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Fixes `java.lang.IllegalArgumentException: No Operation named [init_all_tables] in the Graph` when model needs to be deserialized. The deserialization is skipped when the modelis already loaded (so it will only appear on the worker nodes and not the driver) GPT2 does not contain tables and so does not require this command.
Pull Request Test Coverage Report for Build 7942207273Details
💛 - Coveralls |
maziyarpanahi
approved these changes
Feb 19, 2024
maziyarpanahi
merged commit b148e79
into
JohnSnowLabs:release/530-release-candidate
6 checks passed
maziyarpanahi
added a commit
that referenced
this pull request
Feb 27, 2024
…date * fixed all sbt warnings * remove file system url prefix (#14132) * SPARKNLP-942: MPNet Classifiers (#14147) * SPARKNLP-942: MPNetForSequenceClassification * SPARKNLP-942: MPNetForQuestionAnswering * SPARKNLP-942: MPNet Classifiers Documentation * Restore RobertaforQA bugfix * adding import notebook + changing default model + adding onnx support (#14158) * Sparknlp 876: Introducing LLAMA2 (#14148) * introducing LLAMA2 * Added option to read model from model path to onnx wrapper * Added option to read model from model path to onnx wrapper * updated text description * LLAMA2 python API * added method to save onnx_data * added position ids * - updated Generate.scala to accept onnx tensors - added beam search support for LLAMA2 * updated max input length * updated python default params changed test to slow test * fixed serialization bug * Doc sim rank as retriever (#14149) * Added retrieval interface to the doc sim rank approach * Added Python interface as retriever in doc sim ranker --------- Co-authored-by: Stefano Lori <s.lori@izicap.com> * 812 implement de berta for zero shot classification annotator (#14151) * adding code * adding notebook for import --------- Co-authored-by: Maziyar Panahi <maziyar.panahi@iscpif.fr> * Add notebook for fine tuning sbert (#14152) * [SPARKNLP-986] Fixing optional input col validations (#14153) * [SPARKNLP-984] Fixing Deberta notebooks URIs (#14154) * SparkNLP 933: Introducing M2M100 : multilingual translation model (#14155) * introducing LLAMA2 * Added option to read model from model path to onnx wrapper * Added option to read model from model path to onnx wrapper * updated text description * LLAMA2 python API * added method to save onnx_data * added position ids * - updated Generate.scala to accept onnx tensors - added beam search support for LLAMA2 * updated max input length * updated python default params changed test to slow test * fixed serialization bug * Added Scala code for M2M100 * Documentation for scala code * Python API for M2M100 * added more tests for scala * added tests for python * added pretrained * rewording * fixed serialization bug * fixed serialization bug --------- Co-authored-by: Maziyar Panahi <maziyar.panahi@iscpif.fr> * SPARKNLP-985: Add flexible naming for onnx_data (#14165) Some annotators might have different naming schemes for their files. Added a parameter to control this. * Add LLAMA2Transformer and M2M100Transformer to annotator * Add LLAMA2Transformer and M2M100Transformer to ResourceDownloader * bump version to 5.3.0 [skip test] * SPARKNLP-999: Fix remote model loading for some onnx models * used filesystem to check for the onnx_data file (#14169) * [SPARKNLP-940] Adding changes to correctly copy cluster index storage… (#14167) * [SPARKNLP-940] Adding changes to correctly copy cluster index storage when defined * [SPARKNLP-940] Moving local mode control to its right place * [SPARKNLP-940] Refactoring sentToCLuster method * [SPARKNLP-988] Updating EntityRuler documentation (#14168) * [SPARKNLP-940] Adding changes to support storage temp directory (cluster_tmp_dir) * SPARKNLP-1000: Disable init_all_tables for GPT2 (#14177) Fixes `java.lang.IllegalArgumentException: No Operation named [init_all_tables] in the Graph` when model needs to be deserialized. The deserialization is skipped when the modelis already loaded (so it will only appear on the worker nodes and not the driver) GPT2 does not contain tables and so does not require this command. * fixes python documentation (#14172) * revert MarianTransformer.scala * revert HasBatchedAnnotate.scala * revert Preprocessor.scala * Revert ViTClassifier.scala * disable hard exception * Replace hard exception with soft logs (#14179) This reverts commit eb91fde. * move the example from root to examples/ [skip test] * Cleanup some code [skip test] * Update onnxruntime to 1.17.0 [skip test] * Fix M2M100 default model's name [skip test] * Update docs [run doc] * Update Scala and Python APIs --------- Co-authored-by: ahmedlone127 <ahmedlone127@gmail.com> Co-authored-by: Jiamao Zheng <jiamaozheng@users.noreply.github.com> Co-authored-by: Devin Ha <33089471+DevinTDHa@users.noreply.github.com> Co-authored-by: Prabod Rathnayaka <prabod@rathnayaka.me> Co-authored-by: Stefano Lori <wolliq@users.noreply.github.com> Co-authored-by: Stefano Lori <s.lori@izicap.com> Co-authored-by: Danilo Burbano <37355249+danilojsl@users.noreply.github.com> Co-authored-by: Devin Ha <t.ha@tu-berlin.de> Co-authored-by: Danilo Burbano <danilo@johnsnowlabs.com> Co-authored-by: github-actions <action@github.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
This PR fixes
java.lang.IllegalArgumentException: No Operation named [init_all_tables] in the Graphwhen model needs to be deserialized (e.g. broadcast to worker nodes). The deserialization is skipped when the model is already loaded (so the issue will only appear on the worker nodes and not the driver)GPT2 does not contain tables and so does not require this command. Moreover, this is a legacy feature of TF 1.x models.
Motivation and Context
Resolves #14110.
How Has This Been Tested?
Tests for GPT2 are passing on a cluster.
Types of changes
Checklist: