Youngsik Yang comments

Results 12 comments of


                                            Youngsik Yang

[Bug] TVM FFI encountered a Segfault when calling the invoke_stateful function

I've confirmed that it works after making some changes. I'm going to make the commit soon.

[Bug] [FRONTEND][ONNX] Error converting operator LayerNormalization: tvm.error.InternalError: Check failed: (n.defined()) is false: Found null pointer node while traversing AST.

I've just checked and saw that this error doesn't occur after some changes I made to the from_onnx() code. I'll double check and modify the code soon.

[Bug] [FRONTEND][ONNX] Error converting operator LayerNormalization: tvm.error.InternalError: Check failed: (n.defined()) is false: Found null pointer node while traversing AST.

The fix has been merged into the main branch.

[Bug] [FRONTEND][ONNX] fails to convert a valid onnx model due to the wrong process of DepthToSpace operator: Check failed: (expr->struct_info_.defined()) is false: The struct_info is not populated, check if you have normalized the expr

I’ve fixed the root cause and will commit shortly.

[Relax][ONNX][Transform] Add mode choice, new mode, and warning for take()

@Hzfengsy I've rebased the code according to the newly introduced `reflection` mechanism.

[Bug] How to migrate from te.create_schedule and auto_scheduler to TVM v0.20’

### What changed in TVM 0.20 `te.create_schedule()` (the old TE schedule object) is now deprecated. The [new workflow](https://tvm.apache.org/docs/arch/index.html#overall-flow) is as follows: ![Image](https://github.com/user-attachments/assets/1dbf255d-f701-4269-8fa5-3c40f64fcbe5) The first step is to create an **IRModule**...

[Bug] Inconcistency after meta schedule for the conv2d_hwcn operator

Hi, I checked and compared the values of each output and confirmed that all gaps are within 1 ULP. But in your result log, the gaps seem quite big (quite...

[Bug] inconsistent results for the CPU and CUDA targets.

Hi, I've just found the root causes and confirmed that both cpu and gpu produce the same result. I'll upload code changes very soon.

[Bug] MetaSchedule lead to a significant difference in the reference results for the log_softmax operator

Hi, same as issue #17977 I checked that all value gaps are within 2 ULPs, in other words, no bug

[Bug] LegalizeOps fails on valid onnx model: AssertionError: emit_te now only supports Tensor that has ShapeExpr shape

I think I've fixed the bug. I'll commit soon.