> If I'm interpreting your message correctly, your concern is that Relax's 
> current approach towards development does not yet utilize any of these task 
> tracking mechanisms.

Apologies, I'll clarify, there is no need for Relax to use any of TVMs 
development process, as it's an exploratory project in isolation and any 
pre-upstreaming part of the Relax roadmap is more useful to the Relax 
developers and should be maintained in the Relax repo.

Therefore, I would suggest that a Relax Roadmap within TVM would instead begin 
part way into:

> A1: In the Relax Roadmap RFC, clarify the scope of the roadmap

Out of scope would be `RFCs that the Relax community has made for Relax` as 
they are made purposefully for the development of Relax rather than TVM. TVM 
integration would begin with `The eventual (but yet-to-be-defined) upstreaming 
process of Relax functionalities to TVM mainline` which I believe is then:

> A2: Propose a set of RFCs for upstreaming Relax functionalities to TVM 
> mainline

As that is when the integration work begins, this would then follow the normal 
TVM workflow for integrating a new feature into TVM with review within the TVM 
community for the RFCs and PRs. Once the initial integration work is concluded 
I think you're proposing for Relax work to continue within TVM, therefore the 
roadmapping required would be for `The ongoing efforts to improve Relax 
capabilities in TVM mainline, after initial upstreaming.` which is in line with:

> A3: Propose to move Relax development to TVM mainline

At which point we can figure out whether it still needs its own roadmap or 
whether it becomes a task on the Graph Computations and High-Level 
Optimizations roadmap. The Relax community would also be free to continue 
working in the fork and integrating into TVM as it makes sense to do so.

-- 
Reply to this email directly or view it on GitHub:
https://github.com/apache/tvm-rfcs/pull/69#issuecomment-1122263562
You are receiving this because you are subscribed to this thread.

Message ID: <apache/tvm-rfcs/pull/69/c1122263...@github.com>

Reply via email to