---
base_model:
- meta-llama/Meta-Llama-3-8B
- ytu-ce-cosmos/Turkish-Llama-8b-Instruct-v0.1
- ytu-ce-cosmos/Turkish-Llama-8b-DPO-v0.1
tags:
- merge
- mergekit
- lazymergekit
- meta-llama/Meta-Llama-3-8B
- ytu-ce-cosmos/Turkish-Llama-8b-Instruct-v0.1
- ytu-ce-cosmos/Turkish-Llama-8b-DPO-v0.1
---

# Turkish-Llama3-8B-Merged

Turkish-Llama3-8B-Merged is a merge of the following models using [LazyMergekit](https://colab.research.google.com/drive/1obulZ1ROXHjYLn6PPZJwRR6GzgQogxxb?usp=sharing):
* [meta-llama/Meta-Llama-3-8B](https://huggingface.co/meta-llama/Meta-Llama-3-8B)
* [ytu-ce-cosmos/Turkish-Llama-8b-Instruct-v0.1](https://huggingface.co/ytu-ce-cosmos/Turkish-Llama-8b-Instruct-v0.1)
* [ytu-ce-cosmos/Turkish-Llama-8b-DPO-v0.1](https://huggingface.co/ytu-ce-cosmos/Turkish-Llama-8b-DPO-v0.1)

## 🧩 Configuration

```yaml
base_model: meta-llama/Meta-Llama-3-8B

models:
  - model: meta-llama/Meta-Llama-3-8B  # Base model (retain strong foundational performance)
    parameters:
      weight: 1.0
  - model: ytu-ce-cosmos/Turkish-Llama-8b-Instruct-v0.1  # High-quality Turkish instruction tuning
    parameters:
      weight: 0.7  # Increased weight for better instruction following
  - model: ytu-ce-cosmos/Turkish-Llama-8b-DPO-v0.1  # DPO-tuned for Turkish preferences
    parameters:
      weight: 0.5  # Balanced for alignment

# LoRA Adapters (focused on Turkish language/culture)
adapters:
  - model: Yudum/llama3-lora-turkish  # Core Turkish fine-tuning
    parameters:
      weight: 0.8  # Highest priority (language fundamentals)
  - model: Yudum/Meta-Llama-3-8B-Instruct_tr_wiki_tr_wiki_mix  # Turkish Wikipedia knowledge
    parameters:
      weight: 0.6  # Boost cultural/encyclopedic knowledge
  - model: Kasimyildirim/Qwen2-turkish-comnibed-v5-lora  # Additional Turkish data mix
    parameters:
      weight: 0.4  # Complementary reinforcement

merge_method: ties
parameters:
  density: 0.5  # Slightly lower density to reduce noise
  weight: 0.9   # Stronger weight retention for Turkish features
  normalize: true
dtype: bfloat16
```

## 💻 Usage

```python
!pip install -qU transformers accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "matrixportal/Turkish-Llama3-8B-Merged"
messages = [{"role": "user", "content": "What is a large language model?"}]

tokenizer = AutoTokenizer.from_pretrained(model)
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    torch_dtype=torch.float16,
    device_map="auto",
)

outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
print(outputs[0]["generated_text"])
```