---
license: apache-2.0
base_model: Qwen/Qwen2.5-0.5B-Instruct
tags:
  - language-identification
  - knowledge-distillation
  - dcube
  - qwen2
pipeline_tag: text-generation
---

# DCUBE Language Identifier

Fine-tuned **Qwen2.5-0.5B-Instruct** for multilingual language identification via knowledge distillation.

## Usage

**Paste any text** — the embedded chat template automatically:
1. Applies the DCUBE system prompt
2. Wraps your input with `Identify the language:`

### LM Studio
1. Download the **GGUF** file (or convert from this repo)
2. Load the model — **no manual system prompt needed** if the GGUF includes the chat template
3. Paste text directly, e.g. `Обичам да слушам музика вечер`
4. Set **Temperature = 0**

### Python (llama-cpp)
```bash
py run_gguf.py "Your text here"
```

### Hugging Face Transformers
```python
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("YOUR_REPO_ID", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("YOUR_REPO_ID", trust_remote_code=True)

messages = [{"role": "user", "content": "Le temps est magnifique."}]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
# Template auto-adds system prompt + "Identify the language:" prefix
```

## Supported languages

Arabic, Bulgarian, German, Greek, English, Spanish, French, Hindi, Italian, Japanese, Dutch, Polish, Portuguese, Russian, Swahili, Thai, Turkish, Urdu, Vietnamese, Chinese, Malayalam

## System prompt (embedded in chat template)

```
You are DCUBE Language Identifier, a language identification assistant created by DCUBE Ai (www.dcubeai.com). Given a text, respond with ONLY the language name. Do not include any explanation.
```

## Training

- Teacher: Qwen3-14B (knowledge distillation)
- Method: LoRA SFT on papluca/language-identification + Malayalam samples
- DCUBE Ai · www.dcubeai.com