[
https://issues.apache.org/jira/browse/OPENNLP-1442?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17705996#comment-17705996
]
ASF GitHub Bot commented on OPENNLP-1442:
-----------------------------------------
jzonthemtn commented on code in PR #523:
URL: https://github.com/apache/opennlp/pull/523#discussion_r1150597785
##########
opennlp-dl/README.md:
##########
@@ -4,44 +4,50 @@ This module provides OpenNLP interface implementations for
ONNX models using the
**Important**: This does not provide the ability to train models. Model
training is done outside of OpenNLP. This code provides the ability to use ONNX
models from OpenNLP.
-To build with example models, download the models to the `/src/test/resources`
directory. (These are the exported models described below.)
+Models used in the tests are available in the opennlp evaluation test data.
-```
-
-export OPENNLP_DATA=/tmp/
-mkdir /tmp/dl-doccat /tmp/dl-namefinder
+## NameFinderDL
-# Document categorizer model
-wget https://www.dropbox.com/s/n9uzs8r4xm9rhxb/model.onnx?dl=0 -O
$OPENNLP_DATA/dl-doccat/model.onnx
-wget https://www.dropbox.com/s/aw6yjc68jw0jts6/vocab.txt?dl=0 -O
$OPENNLP_DATA/dl-doccat/vocab.txt
+* Export a Huggingface NER model to ONNX, e.g.:
-# Namefinder model
-wget https://www.dropbox.com/s/zgogq65gs9tyfm1/model.onnx?dl=0 -O
$OPENNLP_DATA/dl-namefinder/model.onnx
-wget https://www.dropbox.com/s/3byt1jggly1dg98/vocab.txt?dl=0 -O
$OPENNLP_DATA/dl-/namefinder/vocab.txt
+```
+python -m transformers.onnx --model=dslim/bert-base-NER --feature
token-classification exported
```
-## TokenNameFinder
+## DocumentCategorizerDL
-* Export a Huggingface NER model to ONNX, e.g.:
+* Export a Huggingface classification (e.g. sentiment) model to ONNX, e.g.:
```
-python -m transformers.onnx --model=dslim/bert-base-NER --feature
token-classification exported
+python -m transformers.onnx
--model=nlptown/bert-base-multilingual-uncased-sentiment --feature
sequence-classification exported
```
-* Copy the exported model to `src/test/resources/namefinder/model.onnx`.
-* Copy the model's
[vocab.txt](https://huggingface.co/dslim/bert-base-NER/tree/main) to
`src/test/resources/namefinder/vocab.txt`.
+## SentenceVectors
-Now you can run the tests in `NameFinderDLTest`.
+* Convert a sentence vectors model to ONNX, e.g.:
-## DocumentCategorizer
-
-* Export a Huggingface classification (e.g. sentiment) model to ONNX, e.g.:
+Install dependencies:
```
-python -m transformers.onnx
--model=nlptown/bert-base-multilingual-uncased-sentiment --feature
sequence-classification exported
+python3 -m pip install optimum onnx onnxruntime
+```
+
+Convert the model:
+
```
Review Comment:
I added those.
> Use ONNX Runtime to support sentence-transformers
> -------------------------------------------------
>
> Key: OPENNLP-1442
> URL: https://issues.apache.org/jira/browse/OPENNLP-1442
> Project: OpenNLP
> Issue Type: Task
> Components: Deep Learning
> Reporter: Jeff Zemerick
> Assignee: Jeff Zemerick
> Priority: Major
>
> Use ONNX Runtime to support sentence-transformers. OpenNLP should be able to
> generate embeddings using an ONNX model.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
