Turkish-LLM-14B-Instruct/README.md

---
language:
- tr
- en
license: apache-2.0
base_model: Qwen/Qwen2.5-14B-Instruct
tags:
- turkish
- qwen2
- instruction-tuned
- sft
- qlora
- tr
- reasoning
- conversational
- low-resource
- turkish-nlp
datasets:
- ogulcanaydogan/Turkish-LLM-v10-Training
pipeline_tag: text-generation
model-index:
- name: Turkish-LLM-14B-Instruct
  results:
  - task:
      type: text-generation
    dataset:
      name: MMLU-TR
      type: custom
    metrics:
    - name: accuracy
      type: acc
      value: 0.5977
---

# Turkish-LLM-14B-Instruct

A Turkish-enhanced 14B model fine-tuned from Qwen2.5-14B-Instruct with QLoRA on 242K Turkish instruction examples.

Part of the [Turkish LLM Family](https://huggingface.co/collections/ogulcanaydogan/turkish-llm-family-69b303b4ef1c36caffca4e94).

## Highlights

- **14B parameters** - strong performance with moderate hardware requirements
- **Outperforms base** on MMLU-TR (+0.30 vs Qwen2.5-14B-Instruct)
- **Live demo** - [Try it on Spaces](https://huggingface.co/spaces/ogulcanaydogan/Turkish-LLM-14B-Chat)
- **GGUF available** - [Q4/Q5/Q8/F16 quantizations](https://huggingface.co/ogulcanaydogan/Turkish-LLM-14B-Instruct-GGUF)

## Benchmark Results

| Benchmark | Base (Qwen2.5-14B) | **Ours** | Delta |
|:---|:---:|:---:|:---:|
| **MMLU-TR** | 0.5947 | **0.5977** | **+0.30** |

## Quick Start

### With Ollama

```bash
ollama run hf.co/ogulcanaydogan/Turkish-LLM-14B-Instruct-GGUF:Q4_K_M
```

### With Transformers

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("ogulcanaydogan/Turkish-LLM-14B-Instruct", torch_dtype="auto", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("ogulcanaydogan/Turkish-LLM-14B-Instruct")

messages = [
    {"role": "system", "content": "Sen yardimci bir Turkce asistansin."},
    {"role": "user", "content": "Yapay zekanin egitim sektorundeki etkilerini acikla."}
]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer([text], return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=512)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
```

## Training Details

| Parameter | Value |
|:---|:---|
| Base Model | Qwen/Qwen2.5-14B-Instruct |
| Method | QLoRA (4-bit NF4) |
| LoRA rank / alpha | 32 / 64 |
| Learning rate | 1e-5 |
| Dataset | 242K Turkish instruction examples |

## Turkish LLM Family

| Model | Size | MMLU-TR | GGUF |
|:---|:---:|:---:|:---|
| [Turkish-LLM-7B](https://huggingface.co/ogulcanaydogan/Turkish-LLM-7B-Instruct) | 7B | - | [Download](https://huggingface.co/ogulcanaydogan/Turkish-LLM-7B-Instruct-GGUF) |
| **[Turkish-LLM-14B](https://huggingface.co/ogulcanaydogan/Turkish-LLM-14B-Instruct)** | **14B** | **0.5977** | [Download](https://huggingface.co/ogulcanaydogan/Turkish-LLM-14B-Instruct-GGUF) |
| [Turkish-LLM-32B](https://huggingface.co/ogulcanaydogan/Turkish-LLM-32B-Instruct) | 32B | 0.6564 | [Download](https://huggingface.co/ogulcanaydogan/Turkish-LLM-32B-Instruct-GGUF) |

## Citation

```bibtex
@misc{aydogan2026turkishllm,
  title={Turkish LLM Family: Open-Source Turkish Language Models},
  author={Ogulcan Aydogan},
  year={2026},
  url={https://huggingface.co/collections/ogulcanaydogan/turkish-llm-family-69b303b4ef1c36caffca4e94}
}
```