mirror of https://github.com/llvm/torch-mlir
15 lines
725 B
Markdown
15 lines
725 B
Markdown
|
# 2021Q4 Roadmap
|
||
|
|
|
||
|
|
**NOTE**: Under construction. Feedback from the
|
||
|
|
[Torch-MLIR RFC](https://discuss.pytorch.org/t/torch-mlir-bridging-pytorch-and-mlir-ecosystems/133151)
|
||
|
|
is expected to influence this!
|
||
|
|
|
||
|
|
1. Demonstrate viability of our solutions on industry-standard workloads:
|
||
|
|
- {inference, training} x {BERT-L, MaskRCNN, ResNet50}
|
||
|
|
1. One industry partner migrates (or begins to migrate) to Torch-MLIR for an e2e
|
||
|
|
flow on their critical path.
|
||
|
|
1. Torch-MLIR is seen/accepted as an "off the shelf" solution for Torch/MLIR
|
||
|
|
interop. This specifically covers "softer" / "quasi-technical" aspects of the
|
||
|
|
project, such as community outreach/recognition, build system integration,
|
||
|
|
testing, CI, packaging, documentation.
|