# Entire Smoketest Used For Scoring?

**URL:** <https://community.drivendata.org/t/entire-smoketest-used-for-scoring/11414>\
**Category:** Children’s Speech Recognition Challenge\
**Created:** [March 14, 2026, 6:04pm UTC](https://community.drivendata.org/t/entire-smoketest-used-for-scoring/11414 "2026-03-14T18:04:55Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![mitchelld12345](https://avatars.discourse-cdn.com/v4/letter/m/c4cdca/32.png) [@mitchelld12345](https://community.drivendata.org/u/mitchelld12345)\
**Post date:** [March 14, 2026, 6:04pm UTC](https://community.drivendata.org/t/entire-smoketest-used-for-scoring/11414/1 "2026-03-14T18:04:56Z")

</div>

I’m using a Blackwell GPU for finetuning and noticing a bit of a gap (\> 1 WER) when testing locally using the Docker container (same as runtime’s example, just with CUDA 12.8). Is the entire 9k smoketest scored when it is ran through the Docker container? Trying to determine if the gap is due to only part of the smoketest being scored or GPU architecture difference.

---

<div class="post-metadata">

**Author:** ![cszc](https://avatars.discourse-cdn.com/v4/letter/c/c68b51/32.png) [@cszc](https://community.drivendata.org/u/cszc)\
**Post date:** [March 14, 2026, 10:05pm UTC](https://community.drivendata.org/t/entire-smoketest-used-for-scoring/11414/2 "2026-03-14T22:05:13Z")

</div>

@mitchelld12345 yes, the entire smoketest is scored when run through the Docker container.