# Sampling with FSMTForConditionalGeneration

**URL:** <https://discuss.huggingface.co/t/sampling-with-fsmtforconditionalgeneration/2939>\
**Category:** 🤗Transformers\
**Created:** [December 23, 2020, 1:28pm UTC](https://discuss.huggingface.co/t/sampling-with-fsmtforconditionalgeneration/2939 "2020-12-23T13:28:20Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![jyoo](https://avatars.discourse-cdn.com/v4/letter/j/a9a28c/32.png) [@jyoo](https://discuss.huggingface.co/u/jyoo)\
**Post date:** [December 23, 2020, 1:28pm UTC](https://discuss.huggingface.co/t/sampling-with-fsmtforconditionalgeneration/2939/1 "2020-12-23T13:28:20Z")

</div>

Hi, I’m looking to perform some English to German translation using `FSMTForConditionalGeneration` model (loading pretrained model from `facebook/wmt19-en-de`). The current documentation only mentions decoding using beam search, but I was wondering if it’s also possible to perform sampling for decoding? I read the blog post here [https://huggingface.co/blog/how-to-generate](https://huggingface.co/blog/how-to-generate) and it talks about sampling-based decoding with the `generate` method. But I wasn’t sure if the same is possible with `FSMTForConditionalGeneration` model. Can I also pass `do_sample=True` and `temperature=<value>` to `FSMTForConditionalGeneration`'s `generate` method to perform sampling with specific temperature?

---

<div class="post-metadata">

**Author:** ![valhalla](https://sea2.discourse-cdn.com/hellohellohello/user_avatar/discuss.huggingface.co/valhalla/32/4179_2.png) [@valhalla](https://discuss.huggingface.co/u/valhalla)\
**Post date:** [December 24, 2020, 7:36am UTC](https://discuss.huggingface.co/t/sampling-with-fsmtforconditionalgeneration/2939/2 "2020-12-24T07:36:00Z")

</div>

Yes, `generate` method is common for all generative models so you can do sampling with FSMT as well.