# Adapt Decision Transformer collator to handle evaluation

**URL:** </static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fadapt-decision-transformer-collator-to-handle-evaluation%2F54985%3E%5C
**Category:** 🤗Transformers\
**Created:** [September 14, 2023, 10:17am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fadapt-decision-transformer-collator-to-handle-evaluation%2F54985 "2023-09-14T10:17:55Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![surprisal-optimizer](https://avatars.discourse-cdn.com/v4/letter/s/bbce88/32.png) [@surprisal-optimizer](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Fsurprisal-optimizer)%5C
**Post date:** [September 14, 2023, 10:17am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fadapt-decision-transformer-collator-to-handle-evaluation%2F54985%2F1 "2023-09-14T10:17:55Z")

</div>

Hi, I am referring to this tutorial [notebook](https://github.com/huggingface/blog/blob/main/notebooks/101_train-decision-transformers.ipynb). I was trying to add an evaluation to the training but now I am encountering some difficulties. It seems that in the original code the Collator does also a custom sampling, but this sampling will always be done on the train\_set since at initialization I will use that one, as shown in the image

 ![ab](https://us1.discourse-cdn.com/hellohellohello/original/3X/9/9/99039d47c89015278dc4a6d0231ec9d799d35a18.png)  
My question is: Is there a way to use 2 different collator logics based on if the model is evaluating or training? Another option I think would be using the collator on the 2 datasets before training but I don’t know if it is possible.

---

<div class="post-metadata">

**Author:** ![eyifjin](https://avatars.discourse-cdn.com/v4/letter/e/f14d63/32.png) [@eyifjin](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Feyifjin)%5C
**Post date:** [July 13, 2024, 3:17pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fadapt-decision-transformer-collator-to-handle-evaluation%2F54985%2F2 "2024-07-13T15:17:04Z")

</div>

Any update regarding this?