# T5Model predict \<UNK\>

**URL:** <https://discuss.huggingface.co/t/t5model-predict-unk/24007>\
**Category:** Beginners\
**Created:** [October 5, 2022, 1:57am UTC](https://discuss.huggingface.co/t/t5model-predict-unk/24007 "2022-10-05T01:57:25Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![Moses25](https://avatars.discourse-cdn.com/v4/letter/m/4bbf92/32.png) [@Moses25](https://discuss.huggingface.co/u/Moses25)\
**Post date:** [October 5, 2022, 1:57am UTC](https://discuss.huggingface.co/t/t5model-predict-unk/24007/1 "2022-10-05T01:57:26Z")

</div>

config = T5Config(num\_layers=12,  
num\_decoder\_layers=12,  
pad\_token\_id=0,  
es\_token\_id=VOCAB\_SIZE + 2,  
model\_parallel=True,  
vocab\_size=VOCAB\_SIZE,  
num\_heads=12,  
d\_model=1024,  
d\_kv=64,  
d\_ff=3072,  
decoder\_start\_token\_id= VOCAB\_SIZE + 1,  
)  
model = T5ForConditionalGeneration(config)  
this is my T5 model config.  
epoch is 200 .I use it to train a translation model,  
but the results of prediction is UNK.

How can I use T5ForConditionalGeneration to train my custom model?  
my config is right?