UmutArchery/Bozdogan-7B
Bozdogan-7B is a 7.6 billion parameter Turkish chat AI model developed by Umut Archery. It is fine-tuned on the Qwen 2.5 7B Instruct base model using approximately 10,000 Turkish chat examples. This model is designed for Turkish conversational AI applications and can run directly with vLLM/serverless setups. It is optimized for generating responses in Turkish and maintains its identity as an Umut Archery creation.
Loading preview...
Bozdogan-7B: A Turkish Chat AI Model
Bozdogan-7B is a 7.6 billion parameter, instruction-tuned large language model developed by Umut Archery, specifically designed for Turkish conversational AI. It is built upon the robust Qwen 2.5 7B Instruct base model.
Key Capabilities & Features
- Turkish Language Focus: Fine-tuned with approximately 10,000 Turkish chat examples, making it highly proficient in generating natural and contextually relevant responses in Turkish.
- Self-Awareness: The model is trained to recognize its own identity, stating that it is Bozdogan and was developed by Umut Archery when asked.
- Direct Deployment: Designed as a single, complete model, it is ready for direct deployment with tools like vLLM or serverless environments.
- Apache 2.0 License: Available under the permissive Apache 2.0 license, allowing for broad use and distribution.
Training Details
The model was trained on an NVIDIA A100 80GB GPU, with the main model training taking approximately 40 minutes, plus an additional 6 minutes for identity training. During training, it achieved a train loss of ~0.53 and an evaluation loss of ~0.59, with a token accuracy of ~88%.
Limitations
As a 7B parameter model trained on a dataset of around 10,000 examples, Bozdogan-7B may exhibit limitations in complex reasoning tasks or highly specialized domains. Users should verify information provided by the model for critical applications.