tvm
tvm copied to clipboard
apache
Open deep learning compiler stack for cpu, gpu and specialized accelerators
This PR adds a new operator that supports embedding_bag operator similar to `torch.embedding_bag`. Co-Authored-By: Masahiro Masuda [[email protected]](mailto:[email protected]) @shingjan @yelite cc @blackkker
See discussion in #12644 for context, this test was being silently skipped before and has broken in the meantime. We should fix it and re-enable it cc @Mousius @areusch @gigiblender
This includes a few fixes for the PR commenter bot: * See https://github.com/apache/tvm/pull/12823#issuecomment-1249785243 for an example, the docs bot and tests bot both needed checks / clarification for when CI...
These tests were found to be flaky (intermittently failing on `main` or failed in a PR with unrelated changes). See [the docs](https://github.com/apache/tvm/blob/main/docs/contribute/ci.rst#handling-flaky-failures) for details. ### Tests(s) - `tests/python/contrib/test_ethosu/test_replace_depthwise_conv2d.py::test_depthwise_conv2d_single[trial3]` ``` [2022-09-19T18:44:28.351Z]...
This PR adds schedule and testcase for uint8 layout. There has been some modification to the testcase so that it can support both nhwc-8h2w32c2w-2d and nhwc-8h8w32c-2d layouts for input and...
Prior to this commit, the following code would compile and run without error. This occurs because the typed `Array::insert` calls the untyped `ArrayNode::InitRange`, with no type-checking done before the call....
This PR: - Adds printing function for Var and Type in TIR - Update the traced_object_functor to require type parameter on `set_dispatch` for better code readability. - Add the TracedObject...
The pass `PlanAndUpdateBufferAllocationLocation` seems to have problem when the buffer accessed indices take a loop carried dependency. As an example, ```python @T.prim_func def test(A: T.Buffer[(8, 8), "int32"], B: T.Buffer[(8, 8),...
### Expected behavior Should compile. ### Actual behavior Doesn't. ### Environment TVM: 60cf692a63a22cd2698273c4945f037b4b22474b Source model: an ONNX file
