---
license: bigscience-bloom-rail-1.0
language:
- hu
---

Base Model:
https://huggingface.co/bigscience/bloomz-7b1

---

Model fine-tuned on a real news dataset and optimized for neural news generation.

Note: Hungarian was not in pretraining.

```python
from transformers import AutoModelForSequenceClassification, AutoTokenizer, pipeline

# Load model and tokenizer
tokenizer = AutoTokenizer.from_pretrained('bigscience/bloomz')
model = AutoModelForSequenceClassification.from_pretrained('tum-nlp/neural-news-generator-bloomz-7b1-hu')

# Create the pipeline for neural news generation and set the repetition penalty >1.1 to punish repetition.
generator = pipeline('text-generation',
                      model=model,
                      tokenizer=tokenizer,
                      repetition_penalty=1.2)

# Define the prompt
prompt = "Cím: Ellenzéki politikai akció az ügyészséggel szemben Cikk: Az ügyészség visszautasítja az igazságszolgáltatást ért politikai nyomásgyakorlást – tájékoztatott [EOP]"

# Generate
generator(prompt, max_length=1000, num_return_sequences=1)

```

Trained on 6k datapoints (including all splits) from the Hungarian news dataset:
https://github.com/batubayk/news_datasets