I agree, I had a lot of fun using it for longformer, and I would love to see it used for more NLP application.
> (if it just worked slightly better) I chatted with the TVM folks, @jknight, @jwfromm, @antinucleon, a few weeks ago and they mentioned they will consider providing better support for our use case. Things like a pip installable runtime and better tutorials can make TVM much easier to use. Better support for fp16 and getting GEMM performance closer to PyTorch's are also important to make it practical. --- [Visit Topic](https://discuss.tvm.ai/t/competitive-gemm-matmul-example/5478/4) to respond. You are receiving this because you enabled mailing list mode. To unsubscribe from these emails, [click here](https://discuss.tvm.ai/email/unsubscribe/f9140da5f9f0b30efd8077cc4e93f04ad22ddb01c3cb6b284799f174cc653986).
