Open Vaccummer opened 2 months ago
Triton doens't support V100 very well
Triton doens't support V100 very well
Thanks for reply, and is there any solution or alternative of Triton?
There's a reference implementation in pytorch but would probably be quite a bit slower
When I train the model of dtype fp16 or mixed precision on V100 sxm2 16G, loss.backward() return error:
python: /project/lib/Analysis/Allocation.cpp:40: std::pair<llvm::SmallVector, llvm::SmallVector > mlir::triton::getCvtOrder(mlir::Attribute, mlir::Attribute): Assertion `!(srcMmaLayout && dstMmaLayout) && "Unexpected mma -> mma layout conversion"' failed.