# Precision loss during conversion

**URL:** <https://community.ultralytics.com/t/precision-loss-during-conversion/1441>\
**Category:** Discussion\
**Tags:** bug-fix\
**Created:** [September 5, 2025, 7:40pm UTC](https://community.ultralytics.com/t/precision-loss-during-conversion/1441 "2025-09-05T19:40:43Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![Andrew\_Qian](https://sea1.discourse-cdn.com/flex001/user_avatar/community.ultralytics.com/andrew_qian/32/826_2.png) [@Andrew\_Qian](https://community.ultralytics.com/u/Andrew_Qian)\
**Post date:** [September 5, 2025, 7:40pm UTC](https://community.ultralytics.com/t/precision-loss-during-conversion/1441/1 "2025-09-05T19:40:43Z")

</div>

Does converting a 32 bit model to 16bit lead to less accurate segmentations? I noticed that my instance segmentation model was slightly off despite being a direct conversion from the full 32 model using onnx followed by tensorRT.

---

<div class="post-metadata">

**Author:** ![Toxite](https://sea1.discourse-cdn.com/flex001/user_avatar/community.ultralytics.com/toxite/32/123_2.png) [@Toxite](https://community.ultralytics.com/u/Toxite)\
**Post date:** [September 6, 2025, 4:41am UTC](https://community.ultralytics.com/t/precision-loss-during-conversion/1441/2 "2025-09-06T04:41:05Z")

</div>

Not really. The PyTorch model checkpoint that Ultralytics saves is FP16. The validation in Ultralytics also runs with FP16 precision.

Did you use Ultralytics to convert to TensorRT?