---
language:
- hu
license: mit
tags:
- hungarian
- causal-lm
- llama
- sentencepiece
library_name: transformers
pipeline_tag: text-generation
model-index:
- name: csermely
  results: []
---

# Csermely

**The smallest coherent Hungarian language model.** Part of the [Emese](https://emese.tech) model family.

Csermely is a 190M parameter decoder-only transformer trained exclusively on high-quality Hungarian text. It runs on edge devices and excels in summarization, grammar checking, and tone detection.

## Model Details

| | |
|---|---|
| **Version** | 0.2 |
| **Parameters** | 190.2M |
| **Context length** | 4,096 tokens (YaRN RoPE, 4× factor) |
| **Architecture** | LLaMA-style (decoder-only transformer) |
| **Training context** | 1,024 tokens |
| **Training precision** | bfloat16 (MLX) |
| **Published weights** | float16 |
| **Vocabulary** | 32,000 (SentencePiece Unigram, Hungarian) |
| **Training data** | ~2B tokens of Hungarian text |
| **License** | MIT |

## Architecture

- 16 transformer layers
- 896 hidden dimension
- 14 attention heads
- 2560 FFN intermediate size
- RMSNorm pre-layer normalization
- Rotary positional embeddings (RoPE) with YaRN extension (4× factor, base 1024)
- SwiGLU feed-forward activation
- Tied input/output embeddings

## Tokenizer

Custom 32K vocabulary SentencePiece Unigram tokenizer trained on high-quality Hungarian corpora. ~30% more token-efficient than multilingual tokenizers for Hungarian text.

Available separately: [emese-tech/emese-tokenizer-32k](https://huggingface.co/emese-tech/emese-tokenizer-32k)

## Usage

```python
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("emese-tech/csermely")
model = AutoModelForCausalLM.from_pretrained("emese-tech/csermely")

input_text = "A magyar nyelv"
inputs = tokenizer(input_text, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=100)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

The default generation config uses `temperature=0.7`, `top_p=0.9`, and `repetition_penalty=1.2` to reduce repetitive output.

## Citation

```bibtex
@misc{emese-csermely-2026,
  title={Csermely: A Hungarian Language Model},
  author={Emese Tech},
  year={2026},
  url={https://huggingface.co/emese-tech/csermely}
}
```