# \[µTVM\] Simplifying the Compiler interface

**URL:** <https://discuss.tvm.apache.org/t/tvm-simplifying-the-compiler-interface/8703>\
**Category:** RFC\
**Created:** [December 16, 2020, 6:41pm UTC](https://discuss.tvm.apache.org/t/tvm-simplifying-the-compiler-interface/8703 "2020-12-16T18:41:44Z")\
**Posts on this page:** 1\
**Showing post:** 3

<div class="post-metadata">

**Author:** ![areusch](https://discuss.tvm.apache.org/user_avatar/discuss.tvm.apache.org/areusch/32/2118_2.png) [@areusch](https://discuss.tvm.apache.org/u/areusch)\
**Post date:** [December 16, 2020, 10:32pm UTC](https://discuss.tvm.apache.org/t/tvm-simplifying-the-compiler-interface/8703/3 "2020-12-16T22:32:56Z")

</div>

@manupa-arm

> from the perspective of deployment flow, are these truly required ?

first let’s clarify deployment flow. so far we’ve demonstrated:

1. standalone deployment, graph runtime on device.
2. deployment with an RPC server and host-driven graph runtime

ultimately I presume deployment means #1. from that perspective, you absolutely don’t need a flasher or transport. these are just useful for case #2 and for autotuning.

> As for the debugging, that would be debugging of operator codegen of tvm and is it an expectation for the user to debug them ?

certainly we would hope this would not be the typical use case, but we should design a debug flow that works.

> I think we need to discuss the possibility of a tuner app

could you elaborate more on this? i think we should avoid linking outside of gcc if we can help it, but if there’s good motivation we should discuss it.

---

_[View the full topic](https://discuss.tvm.apache.org/t/tvm-simplifying-the-compiler-interface/8703)._
