# Debug Runtime & Relay Data

**URL:** <https://discuss.tvm.apache.org/t/debug-runtime-relay-data/9276>\
**Category:** Questions\
**Created:** [March 2, 2021, 9:39am UTC](https://discuss.tvm.apache.org/t/debug-runtime-relay-data/9276 "2021-03-02T09:39:52Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![max1996](https://discuss.tvm.apache.org/letter_avatar_proxy/v4/letter/m/3da27b/32.png) [@max1996](https://discuss.tvm.apache.org/u/max1996)\
**Post date:** [March 2, 2021, 9:39am UTC](https://discuss.tvm.apache.org/t/debug-runtime-relay-data/9276/1 "2021-03-02T09:39:52Z")

</div>

Hi,

For a project I want to build a proof of concept, which should be able to collect layer settings and their latencies in its first step.

I found the debug runtime useful to get the latencies of the layers, however, the information about the optimized and fused layers are very sparse, I cannot find anything about kernel sizes, padding, etc.

Is there a simple way to access the relay representation AFTER the optimization of the build process?

Should I use another method to collect the latency data? Is the debug runtime able to collect the latency of BYOC nodes or subgraphs? ( I was unable to test it myself as I could not build TVM with DNNL - might be a limitation of WSL)
