v0.5.3 Patch Update
This is a patch release mainly for supporting CUDA 11.0. Now DGL supports CUDA 11.0 and PyTorch 1.7 on Linux/Windows/Mac.
Other fixes include:
- Performance fix of graph batching. Affect the
dgl.batch
API. - Speedup on graph readout. Affect all APIs under
dgl.readout
. - Speedup in CPU SpMM with sum reducer.
- Performance optimization that removes redundant copies between CPU and GPU.
- Fix
segment_reduce()
ignoring tailing 0 segments. - Fix a crash bug due to unfound attribute.
- Performance optimization in COO-CSR conversion.
- Parallelization in heterogeneous graph format conversion.
- Fix a bug to enable distributed training of RGCN with CPU.
- Numerous documentation fixes.
New examples:
- Sparse embedding for GATNE-T for large graphs.
- LINE.
- SIGN for OGB.
The Chinese user guide has been released for chapter 1 to 4. Further chapters will be released soon.
Full release note: https://github.com/dmlc/dgl/releases/tag/0.5.3
04 December