elifbasboga/qwen2-7b-instruct-lora-tr

Qwen2-7B-Instruct LoRA TR

Bu model, Qwen/Qwen2-7B-Instruct tabanlı olarak, Türkçe diyaloglar üzerine LoRA ve PEFT teknikleriyle fine-tune edilmiştir.
Fine-tune işlemi Google Colab Pro ortamında, A100 GPU üzerinde, bitsandbytes ile quantized (8-bit) olarak gerçekleştirilmiştir.

Model Details

• Base Model: Qwen/Qwen2-7B-Instruct
• Fine-tuned Model: elifbasboga/qwen2-7b-instruct-lora-tr
• Method: LoRA (Parameter-Efficient Fine-Tuning), PEFT
• Language(s): Turkish
• Libraries: transformers, peft, bitsandbytes
• License: Apache 2.0 (base model) – lütfen kendi kullanımına göre belirle!

Model Description

This model is a Turkish conversational AI based on Qwen2-7B-Instruct, fine-tuned with LoRA adapters on custom Turkish dialog dataset.
It is suitable for chatbot, assistant, and Turkish NLP tasks.

Model Sources

• Base Model: Qwen/Qwen2-7B-Instruct
• This Repository

Uses

Direct Use

• Turkish conversational tasks
• Chatbot and assistant applications

Downstream Use

• Can be further fine-tuned for domain-specific Turkish tasks.

Out-of-Scope Use

• Not suitable for tasks outside Turkish language modeling.
• Not recommended for critical applications without further evaluation.

Bias, Risks, and Limitations

• May reflect biases present in the training data (Turkish conversations).
• Outputs may sometimes be off-topic, incorrect, or inappropriate.

Recommendations

• Use responsibly and review outputs, especially in sensitive or production settings.

How to Get Started

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("elifbasboga/qwen2-7b-instruct-lora-tr")
tokenizer = AutoTokenizer.from_pretrained("elifbasboga/qwen2-7b-instruct-lora-tr")

Training Procedure

• Epochs: 3
• Batch Size: 1
• Learning Rate: 2e-4
• Quantization: 8-bit (bitsandbytes)
• Adapter: LoRA

Hardware

• Google Colab Pro, NVIDIA A100 GPU

Software

• Python 3.11
• transformers, peft, bitsandbytes, datasets

Evaluation

• Model was evaluated qualitatively during training via loss metrics.
• For best results, further testing on your own data is recommended.

Environmental Impact

• GPU: NVIDIA A100
• Cloud Provider: Google Colab

Citation

If you use this model, please cite the base model and this repository.

@misc{elifbasboga_qwen2_7b_instruct_lora_tr_2025,
  title={Qwen2-7B-Instruct LoRA Turkish},
  author={elifbasboga},
  year={2025},
  howpublished={\url{https://huggingface.co/elifbasboga/qwen2-7b-instruct-lora-tr}},
}

Model Card Authors

• elifbasboga

Contact

• Hugging Face Profile