tvm
https://github.com/apache/tvm
Python
Open deep learning compiler stack for cpu, gpu and specialized accelerators
Triage Issues!
When you volunteer to triage issues, you'll receive an email each day with a link to an open issue that needs help in this project. You'll also receive instructions on how to triage issues.
Triage Docs!
Receive a documented method or class from your favorite GitHub repos in your inbox every day. If you're really pro, receive undocumented methods or classes and supercharge your commit history.
Python not yet supported3 Subscribers
Add a CodeTriage badge to tvm
Help out
- Issues
- [Relax][LegalizeOps] Fix GQA/MQA attention legalization by repeating KV heads
- [Fix][Relax][Frontend][Torch] Return the input dtype from float16 and bfloat16 matmul and conv
- [Relax][Frontend][Torch] Add amax and amin converters
- [FIX][Relax] Skip ExpandMatmulOfSum when addition broadcasts
- [Fix][Relax] Preserve zero extents in ONNX import and reshape roundtrips
- [Bug] ExpandMatmulOfSum silently miscompiles matmul(x, a+b) when addends broadcast
- [FIX][Relax] Infer call_te output vdevice from keyword arguments
- [Relax][Frontend][TFLite] Add DYNAMIC_UPDATE_SLICE operator mapping
- [Bug] CUDA workspace cleanup aborts on Windows at thread/process exit
- [Bug][Relax][CUDA/NVPTX] `R.median` legalizes to the CPU `tvm.contrib.sort.argsort` extern, which ends up inside a device kernel: `Unresolved call tirx.call_packed` (cuda) / `unknown intrinsic tirx.call_packed` (nvptx)
- Docs
- Python not yet supported