README.md

---
license: apache-2.0
tags:
- merge
- mergekit
- lazymergekit
- Qwen/Qwen2.5-0.5B-Instruct
---

# Model Merging

## abdullah-Qwen-Qwen2.5-0.5B-Instruct-merged-two-same-model-slerp

abdullah-Qwen-Qwen2.5-0.5B-Instruct-merged-two-same-model-slerp is a merge of the following models using [mergekit](https://github.com/cg123/mergekit):
* [Qwen/Qwen2.5-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct)
* [Qwen/Qwen2.5-0.5B-Instruct](https://huggingface.co/Qwen/Qwen2.5-0.5B-Instruct)

## 🧩 Configuration

Note: You nedd to change the layer range based on `number of layers` in the model. For `Qwen/Qwen2.5-0.5B-Instruct` number of layer is 24. The notebook used (see below in reference section
had `layer_range` `[0, 32]`, this is not true for the qwen model I am using here)

```yaml
slices:
  - sources:
      - model: Qwen/Qwen2.5-0.5B-Instruct
        layer_range: [0, 24]
      - model: Qwen/Qwen2.5-0.5B-Instruct
        layer_range: [0, 24]
merge_method: slerp
base_model: Qwen/Qwen2.5-0.5B-Instruct
parameters:
  t:
    - filter: self_attn
      value: [0, 0.5, 0.3, 0.7, 1]
    - filter: mlp
      value: [1, 0.5, 0.7, 0.3, 0]
    - value: 0.5
dtype: bfloat16

```



## Quickstart

Here provides a code snippet with `apply_chat_template` to show you how to load the tokenizer and model and how to generate contents.

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "abdullahalzubaer/abdullah-Qwen-Qwen2.5-0.5B-Instruct-merged-two-same-model-slerp"

model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

prompt = "Give me a short introduction to large language model."
messages = [
    {"role": "system", "content": "You are Qwen, created by Alibaba Cloud. You are a helpful assistant."},
    {"role": "user", "content": prompt}
]
text = tokenizer.apply_chat_template(
    messages,
    tokenize=False,
    add_generation_prompt=True
)
model_inputs = tokenizer([text], return_tensors="pt").to(model.device)

generated_ids = model.generate(
    **model_inputs,
    max_new_tokens=512
)
generated_ids = [
    output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
]

response = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)[0]
```


## Reference:

Used this https://huggingface.co/blog/mlabonne/merge-models for creating the merged model

## Further Reading

[An Introduction to Model Merging for LLMs](https://developer.nvidia.com/blog/an-introduction-to-model-merging-for-llms/)

[Fine-tuning German LLMs with Model Merging and DPO for Improving Customer Support](https://blog.mayflower.de/17424-fine-tuning-german-llm.html)

[Model Merging: Combining Different Fine-Tuned LLMs](https://blog.marvik.ai/2024/06/19/model-merging-combining-different-fine-tuned-llms/)

## Citation

If you find this work helpful, feel free to give the works below a cite and this repo.

```
@misc{qwen2.5,
    title = {Qwen2.5: A Party of Foundation Models},
    url = {https://qwenlm.github.io/blog/qwen2.5/},
    author = {Qwen Team},
    month = {September},
    year = {2024}
}

@article{qwen2,
      title={Qwen2 Technical Report}, 
      author={An Yang and Baosong Yang and Binyuan Hui and Bo Zheng and Bowen Yu and Chang Zhou and Chengpeng Li and Chengyuan Li and Dayiheng Liu and Fei Huang and Guanting Dong and Haoran Wei and Huan Lin and Jialong Tang and Jialin Wang and Jian Yang and Jianhong Tu and Jianwei Zhang and Jianxin Ma and Jin Xu and Jingren Zhou and Jinze Bai and Jinzheng He and Junyang Lin and Kai Dang and Keming Lu and Keqin Chen and Kexin Yang and Mei Li and Mingfeng Xue and Na Ni and Pei Zhang and Peng Wang and Ru Peng and Rui Men and Ruize Gao and Runji Lin and Shijie Wang and Shuai Bai and Sinan Tan and Tianhang Zhu and Tianhao Li and Tianyu Liu and Wenbin Ge and Xiaodong Deng and Xiaohuan Zhou and Xingzhang Ren and Xinyu Zhang and Xipin Wei and Xuancheng Ren and Yang Fan and Yang Yao and Yichang Zhang and Yu Wan and Yunfei Chu and Yuqiong Liu and Zeyu Cui and Zhenru Zhang and Zhihao Fan},
      journal={arXiv preprint arXiv:2407.10671},
      year={2024}
}
```