Tianqi Chen
Tianqi Chen
Ideally we don't want to change FuseOps behavior, since in cases where expressions are intermediate (e.g. intermediate compute include values that contains exprs like n * 4). This is because...
thanks for the contribution, just want to bring some of the context in https://discuss.tvm.apache.org/t/discuss-tvm-core-strategy-for-operator-scheduling-and-tuning/16352 would love to see how we can leverage some of the techniques in MetaSchedule and TensorIR...
Just highlight some of the main comments that are high level - naming: we can probably go with `runtime::Int`, `runtime::Float`, `runtime::Bool` as they are more understandable terms - we should...
This is merged, thanks @Lunderberg !
We just find out some a perf regression introduced by this PR, specifically, during LLM decode function calling overhead(before the first kernel launch) goes up to 1.4 ms. The likely...
maybe it is not too bad to do manual conversion to the array of ir types in python wrapper then leave ffi part lightweight (per your sugggsted option 3) the...
Thanks @chunit-quic for the contribution! I think there are several goals: - G0: Support translation of PT ops, regardless if they appear in normal FX trace for ExportProgram - G1:...
@chunit-quic let me know if it can help address the problems you see, love to working together and get this feature in!
I think update this PR would be sufficient