Closed voe09 closed 3 years ago
Hi @VivekPanyam could you help take a look at the diff. I have rebased the master branch and all CI passed. Thanks!
None of the Linux CI has passed yet, but just unblocked it so it should be running now. Will merge once CI passes :)
Merging #515 (5809213) into master (62d298b) will increase coverage by
0.01%
. The diff coverage is100.00%
.
@@ Coverage Diff @@
## master #515 +/- ##
==========================================
+ Coverage 87.27% 87.29% +0.01%
==========================================
Files 111 111
Lines 7303 7310 +7
==========================================
+ Hits 6374 6381 +7
Misses 929 929
Impacted Files | Coverage Δ | |
---|---|---|
...rc/main/java/com/uber/neuropod/NeuropodTensor.java | 86.84% <100.00%> (+0.35%) |
:arrow_up: |
...rc/main/native/com_uber_neuropod_NeuropodTensor.cc | 72.26% <100.00%> (+1.05%) |
:arrow_up: |
Continue to review full report at Codecov.
Legend - Click here to learn more
Δ = absolute <relative> (impact)
,ø = not affected
,? = missing data
Powered by Codecov. Last update 62d298b...5809213. Read the comment docs.
Summary:
In neuropod JNI, it does not support STRING_TENSOR as model output. The reason is when it tries to create a Java Tensor from model output, it would check the tensor type. Only numerical tensors are supported and these tensors would be converted to ByteBuffer. However, Java NeuropodTensor itself do support STRING_TENSOR. It can get strings from the tensor data itself. Thus we refactor nativeGetBuffer method. This method would assign a null buffer to bytebuffer if it is a string tensor
Test Plan:
Unit test