# Extra eos and sos tokens in model generated output

**URL:** </static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fextra-eos-and-sos-tokens-in-model-generated-output%2F51227%3E%5C
**Category:** Beginners\
**Created:** [August 18, 2023, 12:56pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fextra-eos-and-sos-tokens-in-model-generated-output%2F51227 "2023-08-18T12:56:47Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![ludog](https://avatars.discourse-cdn.com/v4/letter/l/3ab097/32.png) [@ludog](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Fu%2Fludog)%5C
**Post date:** [August 18, 2023, 12:56pm UTC](/static-proxy?url=https%3A%2F%2Fdiscuss.huggingface.co%2Ft%2Fextra-eos-and-sos-tokens-in-model-generated-output%2F51227%2F1 "2023-08-18T12:56:47Z")

</div>

When I call inference manually on a model (in this case kabita-choudhary/finetuned-bart-for-conversation-summary), the generated outputs all start with a load of SOS tokens. Am I calling the model incorrectly?

```auto
>>> from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
>>> import torch

>>> tokenizer = AutoTokenizer.from_pretrained("kabita-choudhary/finetuned-bart-for-conversation-summary")
>>> model = AutoModelForSeq2SeqLM.from_pretrained("kabita-choudhary/finetuned-bart-for-conversation-summary")

>>> input_text = 'Mick: Are you going to be late?\nJohn: Yes, I think so'
>>> tokenized_text = tokenizer(input_text)['input_ids']
>>> input_tokens = torch.tensor([tokenized_text])

>>> output_tokens = model.generate(input_tokens)
>>> tokenizer.batch_decode(output_tokens)

['</s><s><s><s>John is going to be late.</s>']

```