DJL Monthly Update - October 2021

[zachgk]

Released DJL 0.13.0 (Release Notes)

Feature Enhancements

DJL

• [ci] Add publish python engine and serving wlm pipeline (#1260)
• [python] Add benchmark python engine (#1267)
• [onnx] upgrade OnnxRuntime to 1.9.0 and add UINT8 support (#1271)
• [api] Use engine specific split() (#1269)
• Update default image preprocessing (#1268)
• Add resnet101 to modelzoo (#1272)
• [onnx] Update OrtEngine version to 1.9.0 (#1276)
• Update Artifact.toString() (#1273)
• Upgrade AWS SDK to 2.17.54 (#1277)
• Update djl-serving publish with staging repo id. (#1281)
• [api] Add clear() to PairList (#1282)
• [mxnet] Allow load model without parameter file (#1283)
• [api] Add support for ImageServingTranslator (#1285)
• Differ output serialization (#1288)
• Upgrade DJL version from 0.12.0 to 0.13.0 (#1289)
• Update dependency versions to latest (#1293)
• Increase build number to 0.14.0-SNAPSHOT (#1294)
• Upgrade pytorch to 1.9.1 (#1297)
• [tflite] remove native from pom dependency (#1301)
• [ci] publish python engine snapshot build nightly (#1309)
• Create NoBatchifyTranslator (#1243)
• [model-zoo] Make QA model serving ready (#1311)
• [pytorch] make map_location optional (#1312)
• [pytorch] Add map_location support for load model from InputStream (#1314)
• [pytorch] upload precxx11 to s3 for -auto package (#1320)
• [ci] build pytorch jni for different version (#1322)
• [pytorch] Allow use different version of PyTorch native library (#1323)
• Refactor BERT Translator and Tokenizer (#1318)
• [pytorch] make pytorch-native-auto package optional (#1326)
• [pytorch] Implement padding to bert tokenizer (#1328)

DJL Serving

• Update for input/output api change (#14)
• Add maven publish task to python engine (#13)
• [python] remove type hint to be compatible with lower version of python (#15)
• [pyhton] Add PyTorch resnet18 example (#16)
• Refactor model manager inference api (#17)
• [python] accpet null NDArray to convert to PyNDArray
• [serving] avoid hard-coded request size limit (#18)
• Update publish build script for release to staging repo
• Remove -SNAPSHOT for release publish
• [python] Update PyNDArray due to api interface change
• Upgrade to djl 0.13.0
• enable zip distribution
• Upgrade to djl 0.14.0-SNAPSHOT
• Create ServingModel (#19)
• [python] Allow set python executable path (#21)
• Allow test python model locally (#22)
• [python] warn up python model. (#23)
• Add version back to ModelInfo (#27)
• [serving] improve model status with model version support (#25)
• Remove native-auto dependencies (#28)
• Create Workflow for ModelServing (#26)
• [python] Allow custom environment variable per model (#29)

DJL Clojure

• add test for argmax, argmin, sort
• add sqrt func for ndarray
• add stop-gradient
• add multimethod for nd/stack

Documentation & Examples

DJL

• Update README.md (#1261)
• Add missing docs files to site navigation (#1263)
• [doc] fix error in example doc of NDArray.sort (#1266)
• Update pytorch inference optimization document (#1275)
• Add 0.13.0 release to README page (#1291)
• [docs] Add TensorRT and python engine to dependency README. (#1296)
• Fix link on demodocs (#1298)
• Update benchmark README (#1299)

DJL Demos

• upgrade axios to 0.21.2 (#180)
• Minor fixes for docs site (#181)
• clean gradle.properties
• [canary] Add TensorRT and python test
• Fix typo in canary test
• Add python hybrid engine demo (#182)
• [huggingface] Fix repeatedly initializing model issue
• Change qa model to uncased
• Move huggingface demo into huggingface folder
• Replace webcam-capture with opencv.
• Update canary workflow default version to 0.13.0

D2L Java (zh)

• Add gradle project to debug jupyter note book java code
• Remove unnecessary build on push
• Upgrade to djl 0.13.0
• Fix build failure
• Fix broken jupyter notebook

D2L Java

• Add gradle project to debug jupyter note book java code
• Upgrade to djl 0.13.0 (#176)
• Remove unnecessary build on push
• Minor improvement.
• Fix broken jupyter notebook

Bug Fixes

DJL

• Fix Resource closed exception (#1262)
• [api] input property key should be case-insensitive (#1264)
• Fix serving publish workflow (#1265)
• [pytorch] Fix PtNDArray interoperability with other engine (#1270)
• Fix build failure on GPU (#1279)
• Fix set() for Hybrid engine (#1286)
• [tensorrt] check CUDA compute capability. (#1284)
• [api] avoid save arguments passed from url to metadata.json (#1287)
• Fix exception message (#1300)
• [pytorch] Fix pytorch GPU model loading issue (#1302)
• [bom] Fix pytorch snapshot dependency (#1305)
• Fix NDManager is closed bug (#1308)
• [ci] PyTorch 1.9.1 requires glibc 2.27, no longer support AL2 (#1313)
• [ci] fix nightly serving publish (#1315)
• Fix memory in Trainer.checkGradients (#1319)

DJL Serving

• Fix hang when exception in sending output
• [python] Fix bug in get_as_bytes() (#20)
• [python] Fix resnet18 model GPU bug (#24)