# Train large models on large datasets by parts

**URL:** </static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Ftrain-large-models-on-large-datasets-by-parts%2F5730%3E%5C
**Category:** Beginners\
**Created:** [April 24, 2021, 7:03am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Ftrain-large-models-on-large-datasets-by-parts%2F5730 "2021-04-24T07:03:41Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Constantin](https://avatars.discourse-cdn.com/v4/letter/c/74df32/32.png) [@Constantin](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2FConstantin)%5C
**Post date:** [April 24, 2021, 7:03am UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Ftrain-large-models-on-large-datasets-by-parts%2F5730%2F1 "2021-04-24T07:03:41Z")

</div>

Hi

It is not possible to train large xlm-roberta model on large datasets for token classification due to GPU memory using run\_ner.py

But what if I split my dataset into parts? Is it okay for run\_ner.py script? I just think that result will be worse.